From 41d5c5fd50eee9f80e7de0963cf9cfaa55218c38 Mon Sep 17 00:00:00 2001 From: bghira Date: Mon, 3 Aug 2026 13:17:11 -0600 Subject: [PATCH 1/2] Document model license statuses --- README.es.md | 70 +++++++++++++++++++++++++++++++------------------ README.hi.md | 70 +++++++++++++++++++++++++++++++------------------ README.ja.md | 70 +++++++++++++++++++++++++++++++------------------ README.md | 70 +++++++++++++++++++++++++++++++------------------ README.pt-BR.md | 70 +++++++++++++++++++++++++++++++------------------ README.zh.md | 70 +++++++++++++++++++++++++++++++------------------ 6 files changed, 270 insertions(+), 150 deletions(-) diff --git a/README.es.md b/README.es.md index 7a313b804..4ec1c5d53 100644 --- a/README.es.md +++ b/README.es.md @@ -81,31 +81,51 @@ Para detalles de despliegue, consulta la [Guía Enterprise](/documentation/exper ### Compatibilidad de arquitectura de modelos -| Modelo | Parámetros | PEFT LoRA | Lycoris | Full-Rank | ControlNet | Cuantización | Flow Matching | Codificadores de texto | -|--------|------------|-----------|---------|-----------|------------|--------------|---------------|------------------------| -| **Stable Diffusion XL** | 3.5B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L/G | -| **Stable Diffusion 3** | 2B-8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L/G + T5-XXL | -| **Flux.1** | 12B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL | -| **Flux.2** | 32B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | Mistral-3 Small | -| **Ideogram 4** | 9B | ✓ | ✓ | ✓* | ✗ | fp8/nf4 | ✓ | Qwen3-VL | -| **ACE-Step** | 3.5B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HeartMuLa** | 3B | ✓ | ✓ | ✓* | ✗ | int8 | ✗ | Ninguno | -| **Chroma 1** | 8.9B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | T5-XXL | -| **Auraflow** | 6.8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | UMT5-XXL | -| **PixArt Sigma** | 0.6B-0.9B | ✗ | ✓ | ✓ | ✓ | int8 | ✗ | T5-XXL | -| **Sana** | 0.6B-4.8B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2-2B | -| **Lumina2** | 2B | ✓ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2 | -| **Kwai Kolors** | 5B | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ChatGLM-6B | -| **LTX Video** | 5B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **LTX Video 2** | 19B | ✓ | ✓ | ✓* | ✗ | int8/fp8 | ✓ | Gemma3 | -| **Wan Video** | 1.3B-14B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HiDream** | 17B (8.5B MoE) | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL + Llama | -| **Cosmos2** | 2B-14B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | T5-XXL | -| **OmniGen** | 3.8B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **Qwen Image** | 20B | ✓ | ✓ | ✓* | ✗ | int8/nf4 (req.) | ✓ | T5-XXL | -| **SD 1.x/2.x (Legacy)** | 0.9B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L | - -*✓ = Compatible, ✗ = No compatible, * = Requiere DeepSpeed para entrenamiento full-rank* +SimpleTuner es compatible con las siguientes familias de modelos. El soporte detallado de funciones de entrenamiento está en la [Guía Quickstart](/documentation/QUICKSTART.es.md). + +| Modelo | Parámetros | Licencia | Uso comercial | +| --- | --- | --- | --- | +| **ACE-Step** | 3.5B | Apache-2.0 | Sí | +| **Anima** | No especificado | CircleStone Labs Non-Commercial License v1.2 | No (modelo); salidas permitidas | +| **Auraflow** | 6B | Apache-2.0 | Sí | +| **Boogu-Image** | No especificado | Apache-2.0 | Sí | +| **Chroma 1** | 8.9B | Apache-2.0 | Sí | +| **Cosmos2** | 2B-14B | NVIDIA Open Model License | Sí | +| **Cosmos3** | 16B-65B | OpenMDW-1.1 | Sí | +| **DeepFloyd IF** | 0.4B-4.3B stages | DeepFloyd IF License | Abandonware | +| **ERNIE-Image** | No especificado | Apache-2.0 | Sí | +| **Flux.1** | 8B-12B | Apache-2.0 (schnell); FLUX.1 [dev] Non-Commercial License (dev/Kontext) | Mixto por checkpoint | +| **Flux.2** | 4B-32B | Apache-2.0 (klein 4B); FLUX Non-Commercial License (dev/klein 9B) | Mixto por checkpoint | +| **HeartMuLa** | 3B | No especificada en SimpleTuner | Ver términos upstream | +| **HiDream** | 17B (8.5B MoE) | MIT | Sí | +| **Hunyuan Video** | 8.3B | AGPL-3.0 | Sí (copyleft) | +| **Ideogram 4** | 9B | Ideogram 4 Non-Commercial | No | +| **Kandinsky 5.0 Image** | 6B (lite) | MIT | Sí | +| **Kandinsky 5.0 Video** | 2B lite, 19B pro | MIT | Sí | +| **Kwai Kolors** | 2.7B | Apache-2.0 | Abandonware | +| **Krea2** | No especificado | Krea 2 Community License | Sí (menos de $1M en ingresos; salvaguardas requeridas) | +| **LongCat Image** | 6B | Apache-2.0 | Sí | +| **LongCat Video** | 13.6B | MIT | Sí | +| **LTX Video** | ~2.5B | Apache-2.0 | Sí | +| **LTX Video 2** | 19B | Apache-2.0 | Sí | +| **Lumina2** | 2B | Apache-2.0 | Sí | +| **Mage-Flow** | 4B | MIT | Sí | +| **OmniGen** | 3.8B | MIT | Sí | +| **PixArt Sigma** | 0.6B-0.9B | OpenRAIL++ | Sí (restringido) | +| **Qwen Image** | 20B | Apache-2.0 | Sí | +| **Sana** | 0.6B-4.8B | Apache-2.0 | Sí | +| **Sana Video** | 2B | Apache-2.0 | Sí | +| **SD 1.x/2.x (Legacy)** | 0.9B | OpenRAIL++ | Sí (restringido) | +| **Stable Diffusion 3** | 2B-8B | Stability AI Community License | Sí (menos de $1M en ingresos) | +| **Stable Diffusion XL** | 3.5B | CreativeML OpenRAIL-M | Sí (restringido) | +| **Stable Cascade (Stage C)** | 1B, 3.6B prior | No especificada en SimpleTuner | Abandonware | +| **Wan Video** | 1.3B-14B | Apache-2.0 | Sí | +| **Wan S2V** | 14B | Apache-2.0 | Sí | +| **Z-Image** | 6B | Apache-2.0 | Sí | +| **Z-Image Omni** | 6B | Apache-2.0 | Sí | +| **ZLab I1** | 3B | MIT | Sí | + +*Los valores de licencia se toman de los helpers de modelos de SimpleTuner cuando están disponibles y de las tarjetas/licencias upstream para entradas que antes no estaban especificadas. `No especificada en SimpleTuner` significa que el helper no nombra una licencia y aquí no se resume ningún término upstream; revisa la tarjeta del modelo upstream antes de usarlo.* ### Técnicas avanzadas de entrenamiento diff --git a/README.hi.md b/README.hi.md index 77ff1b91c..699d5bb07 100644 --- a/README.hi.md +++ b/README.hi.md @@ -81,31 +81,51 @@ SimpleTuner एक पूर्ण मल्टी‑यूज़र प्र ### मॉडल आर्किटेक्चर समर्थन {#model-architecture-support} -| मॉडल | पैरामीटर | PEFT LoRA | Lycoris | फुल-रैंक | ControlNet | क्वांटाइज़ेशन | फ्लो मैचिंग | टेक्स्ट एन्कोडर | -|-------|------------|-----------|---------|-----------|------------|--------------|---------------|---------------| -| **Stable Diffusion XL** | 3.5B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L/G | -| **Stable Diffusion 3** | 2B-8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L/G + T5-XXL | -| **Flux.1** | 12B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL | -| **Flux.2** | 32B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | Mistral-3 Small | -| **Ideogram 4** | 9B | ✓ | ✓ | ✓* | ✗ | fp8/nf4 | ✓ | Qwen3-VL | -| **ACE-Step** | 3.5B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HeartMuLa** | 3B | ✓ | ✓ | ✓* | ✗ | int8 | ✗ | कोई नहीं | -| **Chroma 1** | 8.9B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | T5-XXL | -| **Auraflow** | 6.8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | UMT5-XXL | -| **PixArt Sigma** | 0.6B-0.9B | ✗ | ✓ | ✓ | ✓ | int8 | ✗ | T5-XXL | -| **Sana** | 0.6B-4.8B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2-2B | -| **Lumina2** | 2B | ✓ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2 | -| **Kwai Kolors** | 5B | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ChatGLM-6B | -| **LTX Video** | 5B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **LTX Video 2** | 19B | ✓ | ✓ | ✓* | ✗ | int8/fp8 | ✓ | Gemma3 | -| **Wan Video** | 1.3B-14B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HiDream** | 17B (8.5B MoE) | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL + Llama | -| **Cosmos2** | 2B-14B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | T5-XXL | -| **OmniGen** | 3.8B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **Qwen Image** | 20B | ✓ | ✓ | ✓* | ✗ | int8/nf4 (req.) | ✓ | T5-XXL | -| **SD 1.x/2.x (Legacy)** | 0.9B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L | - -*✓ = समर्थित, ✗ = समर्थित नहीं, * = फुल‑रैंक प्रशिक्षण के लिए DeepSpeed आवश्यक* +SimpleTuner निम्नलिखित मॉडल families का समर्थन करता है। विस्तृत training feature समर्थन [Quickstart Guide](/documentation/QUICKSTART.hi.md) में है। + +| मॉडल | पैरामीटर | लाइसेंस | व्यावसायिक उपयोग | +| --- | --- | --- | --- | +| **ACE-Step** | 3.5B | Apache-2.0 | हाँ | +| **Anima** | निर्दिष्ट नहीं | CircleStone Labs Non-Commercial License v1.2 | नहीं (मॉडल); outputs अनुमत | +| **Auraflow** | 6B | Apache-2.0 | हाँ | +| **Boogu-Image** | निर्दिष्ट नहीं | Apache-2.0 | हाँ | +| **Chroma 1** | 8.9B | Apache-2.0 | हाँ | +| **Cosmos2** | 2B-14B | NVIDIA Open Model License | हाँ | +| **Cosmos3** | 16B-65B | OpenMDW-1.1 | हाँ | +| **DeepFloyd IF** | 0.4B-4.3B stages | DeepFloyd IF License | Abandonware | +| **ERNIE-Image** | निर्दिष्ट नहीं | Apache-2.0 | हाँ | +| **Flux.1** | 8B-12B | Apache-2.0 (schnell); FLUX.1 [dev] Non-Commercial License (dev/Kontext) | checkpoint के अनुसार mixed | +| **Flux.2** | 4B-32B | Apache-2.0 (klein 4B); FLUX Non-Commercial License (dev/klein 9B) | checkpoint के अनुसार mixed | +| **HeartMuLa** | 3B | SimpleTuner में निर्दिष्ट नहीं | upstream शर्तें देखें | +| **HiDream** | 17B (8.5B MoE) | MIT | हाँ | +| **Hunyuan Video** | 8.3B | AGPL-3.0 | हाँ (copyleft) | +| **Ideogram 4** | 9B | Ideogram 4 Non-Commercial | नहीं | +| **Kandinsky 5.0 Image** | 6B (lite) | MIT | हाँ | +| **Kandinsky 5.0 Video** | 2B lite, 19B pro | MIT | हाँ | +| **Kwai Kolors** | 2.7B | Apache-2.0 | Abandonware | +| **Krea2** | निर्दिष्ट नहीं | Krea 2 Community License | हाँ ($1M से कम राजस्व; safeguards आवश्यक) | +| **LongCat Image** | 6B | Apache-2.0 | हाँ | +| **LongCat Video** | 13.6B | MIT | हाँ | +| **LTX Video** | ~2.5B | Apache-2.0 | हाँ | +| **LTX Video 2** | 19B | Apache-2.0 | हाँ | +| **Lumina2** | 2B | Apache-2.0 | हाँ | +| **Mage-Flow** | 4B | MIT | हाँ | +| **OmniGen** | 3.8B | MIT | हाँ | +| **PixArt Sigma** | 0.6B-0.9B | OpenRAIL++ | हाँ (restricted) | +| **Qwen Image** | 20B | Apache-2.0 | हाँ | +| **Sana** | 0.6B-4.8B | Apache-2.0 | हाँ | +| **Sana Video** | 2B | Apache-2.0 | हाँ | +| **SD 1.x/2.x (Legacy)** | 0.9B | OpenRAIL++ | हाँ (restricted) | +| **Stable Diffusion 3** | 2B-8B | Stability AI Community License | हाँ ($1M से कम राजस्व) | +| **Stable Diffusion XL** | 3.5B | CreativeML OpenRAIL-M | हाँ (restricted) | +| **Stable Cascade (Stage C)** | 1B, 3.6B prior | SimpleTuner में निर्दिष्ट नहीं | Abandonware | +| **Wan Video** | 1.3B-14B | Apache-2.0 | हाँ | +| **Wan S2V** | 14B | Apache-2.0 | हाँ | +| **Z-Image** | 6B | Apache-2.0 | हाँ | +| **Z-Image Omni** | 6B | Apache-2.0 | हाँ | +| **ZLab I1** | 3B | MIT | हाँ | + +*लाइसेंस मान उपलब्ध होने पर SimpleTuner model helpers से लिए गए हैं, और पहले अनिर्दिष्ट entries के लिए upstream model cards/licenses से। `SimpleTuner में निर्दिष्ट नहीं` का अर्थ है कि helper कोई license name नहीं देता और यहां upstream terms का सार नहीं दिया गया है; उपयोग से पहले upstream model card देखें.* ### उन्नत प्रशिक्षण तकनीकें {#advanced-training-techniques} diff --git a/README.ja.md b/README.ja.md index 4fa0b109c..c68739a3a 100644 --- a/README.ja.md +++ b/README.ja.md @@ -81,31 +81,51 @@ SimpleTunerには、エンタープライズグレードの機能を備えた完 ### モデルアーキテクチャサポート -| モデル | パラメータ数 | PEFT LoRA | Lycoris | Full-Rank | ControlNet | 量子化 | Flow Matching | テキストエンコーダー | -|-------|------------|-----------|---------|-----------|------------|--------------|---------------|---------------| -| **Stable Diffusion XL** | 3.5B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L/G | -| **Stable Diffusion 3** | 2B-8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L/G + T5-XXL | -| **Flux.1** | 12B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL | -| **Flux.2** | 32B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | Mistral-3 Small | -| **Ideogram 4** | 9B | ✓ | ✓ | ✓* | ✗ | fp8/nf4 | ✓ | Qwen3-VL | -| **ACE-Step** | 3.5B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HeartMuLa** | 3B | ✓ | ✓ | ✓* | ✗ | int8 | ✗ | なし | -| **Chroma 1** | 8.9B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | T5-XXL | -| **Auraflow** | 6.8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | UMT5-XXL | -| **PixArt Sigma** | 0.6B-0.9B | ✗ | ✓ | ✓ | ✓ | int8 | ✗ | T5-XXL | -| **Sana** | 0.6B-4.8B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2-2B | -| **Lumina2** | 2B | ✓ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2 | -| **Kwai Kolors** | 5B | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ChatGLM-6B | -| **LTX Video** | 5B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **LTX Video 2** | 19B | ✓ | ✓ | ✓* | ✗ | int8/fp8 | ✓ | Gemma3 | -| **Wan Video** | 1.3B-14B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HiDream** | 17B (8.5B MoE) | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL + Llama | -| **Cosmos2** | 2B-14B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | T5-XXL | -| **OmniGen** | 3.8B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **Qwen Image** | 20B | ✓ | ✓ | ✓* | ✗ | int8/nf4 (req.) | ✓ | T5-XXL | -| **SD 1.x/2.x (Legacy)** | 0.9B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L | - -*✓ = サポート, ✗ = 非サポート, * = Full-rankトレーニングにDeepSpeedが必要* +SimpleTunerは以下のモデルファミリーをサポートしています。詳細なトレーニング機能の対応状況は[Quickstartガイド](/documentation/QUICKSTART.ja.md)を参照してください。 + +| モデル | パラメータ数 | ライセンス | 商用利用 | +| --- | --- | --- | --- | +| **ACE-Step** | 3.5B | Apache-2.0 | 可 | +| **Anima** | 未指定 | CircleStone Labs Non-Commercial License v1.2 | 不可（モデル）；出力は可 | +| **Auraflow** | 6B | Apache-2.0 | 可 | +| **Boogu-Image** | 未指定 | Apache-2.0 | 可 | +| **Chroma 1** | 8.9B | Apache-2.0 | 可 | +| **Cosmos2** | 2B-14B | NVIDIA Open Model License | 可 | +| **Cosmos3** | 16B-65B | OpenMDW-1.1 | 可 | +| **DeepFloyd IF** | 0.4B-4.3B stages | DeepFloyd IF License | Abandonware | +| **ERNIE-Image** | 未指定 | Apache-2.0 | 可 | +| **Flux.1** | 8B-12B | Apache-2.0 (schnell); FLUX.1 [dev] Non-Commercial License (dev/Kontext) | checkpointごとに異なる | +| **Flux.2** | 4B-32B | Apache-2.0 (klein 4B); FLUX Non-Commercial License (dev/klein 9B) | checkpointごとに異なる | +| **HeartMuLa** | 3B | SimpleTunerでは未指定 | 上流の条件を確認 | +| **HiDream** | 17B (8.5B MoE) | MIT | 可 | +| **Hunyuan Video** | 8.3B | AGPL-3.0 | 可（copyleft） | +| **Ideogram 4** | 9B | Ideogram 4 Non-Commercial | 不可 | +| **Kandinsky 5.0 Image** | 6B (lite) | MIT | 可 | +| **Kandinsky 5.0 Video** | 2B lite, 19B pro | MIT | 可 | +| **Kwai Kolors** | 2.7B | Apache-2.0 | Abandonware | +| **Krea2** | 未指定 | Krea 2 Community License | 可（年収100万米ドル未満；安全対策必須） | +| **LongCat Image** | 6B | Apache-2.0 | 可 | +| **LongCat Video** | 13.6B | MIT | 可 | +| **LTX Video** | ~2.5B | Apache-2.0 | 可 | +| **LTX Video 2** | 19B | Apache-2.0 | 可 | +| **Lumina2** | 2B | Apache-2.0 | 可 | +| **Mage-Flow** | 4B | MIT | 可 | +| **OmniGen** | 3.8B | MIT | 可 | +| **PixArt Sigma** | 0.6B-0.9B | OpenRAIL++ | 可（制限あり） | +| **Qwen Image** | 20B | Apache-2.0 | 可 | +| **Sana** | 0.6B-4.8B | Apache-2.0 | 可 | +| **Sana Video** | 2B | Apache-2.0 | 可 | +| **SD 1.x/2.x (Legacy)** | 0.9B | OpenRAIL++ | 可（制限あり） | +| **Stable Diffusion 3** | 2B-8B | Stability AI Community License | 可（年収100万米ドル未満） | +| **Stable Diffusion XL** | 3.5B | CreativeML OpenRAIL-M | 可（制限あり） | +| **Stable Cascade (Stage C)** | 1B, 3.6B prior | SimpleTunerでは未指定 | Abandonware | +| **Wan Video** | 1.3B-14B | Apache-2.0 | 可 | +| **Wan S2V** | 14B | Apache-2.0 | 可 | +| **Z-Image** | 6B | Apache-2.0 | 可 | +| **Z-Image Omni** | 6B | Apache-2.0 | 可 | +| **ZLab I1** | 3B | MIT | 可 | + +*ライセンス値は、利用可能な場合はSimpleTunerのモデルヘルパーから、以前未指定だった項目は上流のモデルカード/ライセンスから取得しています。`SimpleTunerでは未指定`は、ヘルパーがライセンス名を持たず、ここでも上流条件を要約していないことを意味します。使用前に上流モデルカードを確認してください。* ### 高度なトレーニング技術 diff --git a/README.md b/README.md index b38d7cf3c..3a0e88972 100644 --- a/README.md +++ b/README.md @@ -81,31 +81,51 @@ For deployment details, see the [Enterprise Guide](/documentation/experimental/s ### Model Architecture Support -| Model | Parameters | PEFT LoRA | Lycoris | Full-Rank | ControlNet | Ref Inputs | Quantization | Flow Matching | Text Encoders | -|-------|------------|-----------|---------|-----------|------------|------------|--------------|---------------|---------------| -| **Stable Diffusion XL** | 3.5B | ✓ | ✓ | ✓ | ✓ | ✗ | int8/nf4 | ✗ | CLIP-L/G | -| **Stable Diffusion 3** | 2B-8B | ✓ | ✓ | ✓* | ✓ | ✗ | int8/fp8/nf4 | ✓ | CLIP-L/G + T5-XXL | -| **Flux.1** | 12B | ✓ | ✓ | ✓* | ✓ | ✓ (Kontext) | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL | -| **Flux.2** | 32B | ✓ | ✓ | ✓* | ✗ | ✓ opt | int8/fp8/nf4 | ✓ | Mistral-3 Small | -| **Ideogram 4** | 9B | ✓ | ✓ | ✓* | ✗ | ✗ | fp8/nf4 | ✓ | Qwen3-VL | -| **ACE-Step** | 3.5B | ✓ | ✓ | ✓* | ✗ | ✗ | int8 | ✓ | UMT5 | -| **HeartMuLa** | 3B | ✓ | ✓ | ✓* | ✗ | ✗ | int8 | ✗ | None | -| **Chroma 1** | 8.9B | ✓ | ✓ | ✓* | ✗ | ✗ | int8/fp8/nf4 | ✓ | T5-XXL | -| **Auraflow** | 6.8B | ✓ | ✓ | ✓* | ✓ | ✗ | int8/fp8/nf4 | ✓ | UMT5-XXL | -| **PixArt Sigma** | 0.6B-0.9B | ✗ | ✓ | ✓ | ✓ | ✗ | int8 | ✗ | T5-XXL | -| **Sana** | 0.6B-4.8B | ✗ | ✓ | ✓ | ✗ | ✗ | int8 | ✓ | Gemma2-2B | -| **Lumina2** | 2B | ✓ | ✓ | ✓ | ✗ | ✗ | int8 | ✓ | Gemma2 | -| **Kwai Kolors** | 5B | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ChatGLM-6B | -| **LTX Video** | 5B | ✓ | ✓ | ✓ | ✗ | ✓ I2V | int8/fp8 | ✓ | T5-XXL | -| **LTX Video 2** | 19B | ✓ | ✓ | ✓* | ✗ | ✓ opt | int8/fp8 | ✓ | Gemma3 | -| **Wan Video** | 1.3B-14B | ✓ | ✓ | ✓* | ✗ | ✗ | int8 | ✓ | UMT5 | -| **HiDream** | 17B (8.5B MoE) | ✓ | ✓ | ✓* | ✓ | ✗ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL + Llama | -| **Cosmos2** | 2B-14B | ✗ | ✓ | ✓ | ✗ | ✗ | int8 | ✓ | T5-XXL | -| **OmniGen** | 3.8B | ✓ | ✓ | ✓ | ✗ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **Qwen Image** | 20B | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | int8/nf4 (req.) | ✓ | T5-XXL | -| **SD 1.x/2.x (Legacy)** | 0.9B | ✓ | ✓ | ✓ | ✓ | ✗ | int8/nf4 | ✗ | CLIP-L | - -*✓ = Supported, ✗ = Not supported, * = Requires DeepSpeed for full-rank training, Ref Inputs marks existing reference/edit/I2V conditioning paths only* +SimpleTuner supports the following model families. Detailed training feature support lives in the [Quickstart Guide](/documentation/QUICKSTART.md#feature-compatibility). + +| Model | Parameters | License | Commercial use | +| --- | --- | --- | --- | +| **ACE-Step** | 3.5B | Apache-2.0 | Yes | +| **Anima** | Not specified | CircleStone Labs Non-Commercial License v1.2 | No (model); outputs allowed | +| **Auraflow** | 6B | Apache-2.0 | Yes | +| **Boogu-Image** | Not specified | Apache-2.0 | Yes | +| **Chroma 1** | 8.9B | Apache-2.0 | Yes | +| **Cosmos2** | 2B-14B | NVIDIA Open Model License | Yes | +| **Cosmos3** | 16B-65B | OpenMDW-1.1 | Yes | +| **DeepFloyd IF** | 0.4B-4.3B stages | DeepFloyd IF License | Abandonware | +| **ERNIE-Image** | Not specified | Apache-2.0 | Yes | +| **Flux.1** | 8B-12B | Apache-2.0 (schnell); FLUX.1 [dev] Non-Commercial License (dev/Kontext) | Mixed by checkpoint | +| **Flux.2** | 4B-32B | Apache-2.0 (klein 4B); FLUX Non-Commercial License (dev/klein 9B) | Mixed by checkpoint | +| **HeartMuLa** | 3B | Not specified in SimpleTuner | See upstream terms | +| **HiDream** | 17B (8.5B MoE) | MIT | Yes | +| **Hunyuan Video** | 8.3B | AGPL-3.0 | Yes (copyleft) | +| **Ideogram 4** | 9B | Ideogram 4 Non-Commercial | No | +| **Kandinsky 5.0 Image** | 6B (lite) | MIT | Yes | +| **Kandinsky 5.0 Video** | 2B lite, 19B pro | MIT | Yes | +| **Kwai Kolors** | 2.7B | Apache-2.0 | Abandonware | +| **Krea2** | Not specified | Krea 2 Community License | Yes (under $1M revenue; safeguards required) | +| **LongCat Image** | 6B | Apache-2.0 | Yes | +| **LongCat Video** | 13.6B | MIT | Yes | +| **LTX Video** | ~2.5B | Apache-2.0 | Yes | +| **LTX Video 2** | 19B | Apache-2.0 | Yes | +| **Lumina2** | 2B | Apache-2.0 | Yes | +| **Mage-Flow** | 4B | MIT | Yes | +| **OmniGen** | 3.8B | MIT | Yes | +| **PixArt Sigma** | 0.6B-0.9B | OpenRAIL++ | Yes (restricted) | +| **Qwen Image** | 20B | Apache-2.0 | Yes | +| **Sana** | 0.6B-4.8B | Apache-2.0 | Yes | +| **Sana Video** | 2B | Apache-2.0 | Yes | +| **SD 1.x/2.x (Legacy)** | 0.9B | OpenRAIL++ | Yes (restricted) | +| **Stable Diffusion 3** | 2B-8B | Stability AI Community License | Yes (under $1M revenue) | +| **Stable Diffusion XL** | 3.5B | CreativeML OpenRAIL-M | Yes (restricted) | +| **Stable Cascade (Stage C)** | 1B, 3.6B prior | Not specified in SimpleTuner | Abandonware | +| **Wan Video** | 1.3B-14B | Apache-2.0 | Yes | +| **Wan S2V** | 14B | Apache-2.0 | Yes | +| **Z-Image** | 6B | Apache-2.0 | Yes | +| **Z-Image Omni** | 6B | Apache-2.0 | Yes | +| **ZLab I1** | 3B | MIT | Yes | + +*License values are taken from SimpleTuner model helpers when available and upstream model cards/licenses for entries that were previously unspecified. `Not specified in SimpleTuner` means the helper does not name a license and no upstream term is summarized here; check the upstream model card before use.* ### Advanced Training Techniques diff --git a/README.pt-BR.md b/README.pt-BR.md index 19b7ca412..a4ccf22aa 100644 --- a/README.pt-BR.md +++ b/README.pt-BR.md @@ -81,31 +81,51 @@ Para detalhes de deploy, veja o [guia enterprise](/documentation/experimental/se ### Suporte a arquitetura de modelos -| Modelo | Parametros | PEFT LoRA | Lycoris | Full-Rank | ControlNet | Quantizacao | Flow Matching | Text Encoders | -|-------|------------|-----------|---------|-----------|------------|--------------|---------------|---------------| -| **Stable Diffusion XL** | 3.5B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L/G | -| **Stable Diffusion 3** | 2B-8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L/G + T5-XXL | -| **Flux.1** | 12B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL | -| **Flux.2** | 32B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | Mistral-3 Small | -| **Ideogram 4** | 9B | ✓ | ✓ | ✓* | ✗ | fp8/nf4 | ✓ | Qwen3-VL | -| **ACE-Step** | 3.5B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HeartMuLa** | 3B | ✓ | ✓ | ✓* | ✗ | int8 | ✗ | Nenhum | -| **Chroma 1** | 8.9B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | T5-XXL | -| **Auraflow** | 6.8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | UMT5-XXL | -| **PixArt Sigma** | 0.6B-0.9B | ✗ | ✓ | ✓ | ✓ | int8 | ✗ | T5-XXL | -| **Sana** | 0.6B-4.8B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2-2B | -| **Lumina2** | 2B | ✓ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2 | -| **Kwai Kolors** | 5B | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ChatGLM-6B | -| **LTX Video** | 5B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **LTX Video 2** | 19B | ✓ | ✓ | ✓* | ✗ | int8/fp8 | ✓ | Gemma3 | -| **Wan Video** | 1.3B-14B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HiDream** | 17B (8.5B MoE) | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL + Llama | -| **Cosmos2** | 2B-14B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | T5-XXL | -| **OmniGen** | 3.8B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **Qwen Image** | 20B | ✓ | ✓ | ✓* | ✗ | int8/nf4 (req.) | ✓ | T5-XXL | -| **SD 1.x/2.x (Legacy)** | 0.9B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L | - -*✓ = Suportado, ✗ = Nao suportado, * = Requer DeepSpeed para treino full-rank* +SimpleTuner suporta as seguintes familias de modelos. O suporte detalhado a recursos de treinamento fica no [guia Quickstart](/documentation/QUICKSTART.pt-BR.md). + +| Modelo | Parametros | Licenca | Uso comercial | +| --- | --- | --- | --- | +| **ACE-Step** | 3.5B | Apache-2.0 | Sim | +| **Anima** | Nao especificado | CircleStone Labs Non-Commercial License v1.2 | Nao (modelo); outputs permitidos | +| **Auraflow** | 6B | Apache-2.0 | Sim | +| **Boogu-Image** | Nao especificado | Apache-2.0 | Sim | +| **Chroma 1** | 8.9B | Apache-2.0 | Sim | +| **Cosmos2** | 2B-14B | NVIDIA Open Model License | Sim | +| **Cosmos3** | 16B-65B | OpenMDW-1.1 | Sim | +| **DeepFloyd IF** | 0.4B-4.3B stages | DeepFloyd IF License | Abandonware | +| **ERNIE-Image** | Nao especificado | Apache-2.0 | Sim | +| **Flux.1** | 8B-12B | Apache-2.0 (schnell); FLUX.1 [dev] Non-Commercial License (dev/Kontext) | Misto por checkpoint | +| **Flux.2** | 4B-32B | Apache-2.0 (klein 4B); FLUX Non-Commercial License (dev/klein 9B) | Misto por checkpoint | +| **HeartMuLa** | 3B | Nao especificada no SimpleTuner | Ver termos upstream | +| **HiDream** | 17B (8.5B MoE) | MIT | Sim | +| **Hunyuan Video** | 8.3B | AGPL-3.0 | Sim (copyleft) | +| **Ideogram 4** | 9B | Ideogram 4 Non-Commercial | Nao | +| **Kandinsky 5.0 Image** | 6B (lite) | MIT | Sim | +| **Kandinsky 5.0 Video** | 2B lite, 19B pro | MIT | Sim | +| **Kwai Kolors** | 2.7B | Apache-2.0 | Abandonware | +| **Krea2** | Nao especificado | Krea 2 Community License | Sim (menos de $1M em receita; salvaguardas exigidas) | +| **LongCat Image** | 6B | Apache-2.0 | Sim | +| **LongCat Video** | 13.6B | MIT | Sim | +| **LTX Video** | ~2.5B | Apache-2.0 | Sim | +| **LTX Video 2** | 19B | Apache-2.0 | Sim | +| **Lumina2** | 2B | Apache-2.0 | Sim | +| **Mage-Flow** | 4B | MIT | Sim | +| **OmniGen** | 3.8B | MIT | Sim | +| **PixArt Sigma** | 0.6B-0.9B | OpenRAIL++ | Sim (restrito) | +| **Qwen Image** | 20B | Apache-2.0 | Sim | +| **Sana** | 0.6B-4.8B | Apache-2.0 | Sim | +| **Sana Video** | 2B | Apache-2.0 | Sim | +| **SD 1.x/2.x (Legacy)** | 0.9B | OpenRAIL++ | Sim (restrito) | +| **Stable Diffusion 3** | 2B-8B | Stability AI Community License | Sim (menos de $1M em receita) | +| **Stable Diffusion XL** | 3.5B | CreativeML OpenRAIL-M | Sim (restrito) | +| **Stable Cascade (Stage C)** | 1B, 3.6B prior | Nao especificada no SimpleTuner | Abandonware | +| **Wan Video** | 1.3B-14B | Apache-2.0 | Sim | +| **Wan S2V** | 14B | Apache-2.0 | Sim | +| **Z-Image** | 6B | Apache-2.0 | Sim | +| **Z-Image Omni** | 6B | Apache-2.0 | Sim | +| **ZLab I1** | 3B | MIT | Sim | + +*Os valores de licenca vem dos model helpers do SimpleTuner quando disponiveis e dos model cards/licencas upstream para entradas que antes nao estavam especificadas. `Nao especificada no SimpleTuner` significa que o helper nao informa uma licenca e nenhum termo upstream esta resumido aqui; verifique o model card upstream antes de usar.* ### Tecnicas avancadas de treinamento diff --git a/README.zh.md b/README.zh.md index 7e4995b63..e1ce18596 100644 --- a/README.zh.md +++ b/README.zh.md @@ -81,31 +81,51 @@ SimpleTuner 包含完整的多用户训练平台，具有企业级功能——** ### 模型架构支持 -| 模型 | 参数量 | PEFT LoRA | Lycoris | 全秩 | ControlNet | 量化 | Flow Matching | 文本编码器 | -|-------|------------|-----------|---------|-----------|------------|--------------|---------------|---------------| -| **Stable Diffusion XL** | 3.5B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L/G | -| **Stable Diffusion 3** | 2B-8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L/G + T5-XXL | -| **Flux.1** | 12B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL | -| **Flux.2** | 32B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | Mistral-3 Small | -| **Ideogram 4** | 9B | ✓ | ✓ | ✓* | ✗ | fp8/nf4 | ✓ | Qwen3-VL | -| **ACE-Step** | 3.5B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HeartMuLa** | 3B | ✓ | ✓ | ✓* | ✗ | int8 | ✗ | 无 | -| **Chroma 1** | 8.9B | ✓ | ✓ | ✓* | ✗ | int8/fp8/nf4 | ✓ | T5-XXL | -| **Auraflow** | 6.8B | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | UMT5-XXL | -| **PixArt Sigma** | 0.6B-0.9B | ✗ | ✓ | ✓ | ✓ | int8 | ✗ | T5-XXL | -| **Sana** | 0.6B-4.8B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2-2B | -| **Lumina2** | 2B | ✓ | ✓ | ✓ | ✗ | int8 | ✓ | Gemma2 | -| **Kwai Kolors** | 5B | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ChatGLM-6B | -| **LTX Video** | 5B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **LTX Video 2** | 19B | ✓ | ✓ | ✓* | ✗ | int8/fp8 | ✓ | Gemma3 | -| **Wan Video** | 1.3B-14B | ✓ | ✓ | ✓* | ✗ | int8 | ✓ | UMT5 | -| **HiDream** | 17B (8.5B MoE) | ✓ | ✓ | ✓* | ✓ | int8/fp8/nf4 | ✓ | CLIP-L + T5-XXL + Llama | -| **Cosmos2** | 2B-14B | ✗ | ✓ | ✓ | ✗ | int8 | ✓ | T5-XXL | -| **OmniGen** | 3.8B | ✓ | ✓ | ✓ | ✗ | int8/fp8 | ✓ | T5-XXL | -| **Qwen Image** | 20B | ✓ | ✓ | ✓* | ✗ | int8/nf4（必需） | ✓ | T5-XXL | -| **SD 1.x/2.x（旧版）** | 0.9B | ✓ | ✓ | ✓ | ✓ | int8/nf4 | ✗ | CLIP-L | - -*✓ = 支持，✗ = 不支持，* = 全秩训练需要 DeepSpeed* +SimpleTuner 支持以下模型系列。详细的训练功能支持请参阅[快速入门指南](/documentation/QUICKSTART.zh.md)。 + +| 模型 | 参数量 | 许可证 | 商业使用 | +| --- | --- | --- | --- | +| **ACE-Step** | 3.5B | Apache-2.0 | 是 | +| **Anima** | 未指定 | CircleStone Labs Non-Commercial License v1.2 | 否（模型）；输出可商用 | +| **Auraflow** | 6B | Apache-2.0 | 是 | +| **Boogu-Image** | 未指定 | Apache-2.0 | 是 | +| **Chroma 1** | 8.9B | Apache-2.0 | 是 | +| **Cosmos2** | 2B-14B | NVIDIA Open Model License | 是 | +| **Cosmos3** | 16B-65B | OpenMDW-1.1 | 是 | +| **DeepFloyd IF** | 0.4B-4.3B stages | DeepFloyd IF License | Abandonware | +| **ERNIE-Image** | 未指定 | Apache-2.0 | 是 | +| **Flux.1** | 8B-12B | Apache-2.0 (schnell)；FLUX.1 [dev] Non-Commercial License (dev/Kontext) | 按 checkpoint 不同 | +| **Flux.2** | 4B-32B | Apache-2.0 (klein 4B)；FLUX Non-Commercial License (dev/klein 9B) | 按 checkpoint 不同 | +| **HeartMuLa** | 3B | SimpleTuner 中未指定 | 查看上游条款 | +| **HiDream** | 17B (8.5B MoE) | MIT | 是 | +| **Hunyuan Video** | 8.3B | AGPL-3.0 | 是（copyleft） | +| **Ideogram 4** | 9B | Ideogram 4 Non-Commercial | 否 | +| **Kandinsky 5.0 Image** | 6B (lite) | MIT | 是 | +| **Kandinsky 5.0 Video** | 2B lite, 19B pro | MIT | 是 | +| **Kwai Kolors** | 2.7B | Apache-2.0 | Abandonware | +| **Krea2** | 未指定 | Krea 2 Community License | 是（年收入低于 100 万美元；需安全防护） | +| **LongCat Image** | 6B | Apache-2.0 | 是 | +| **LongCat Video** | 13.6B | MIT | 是 | +| **LTX Video** | ~2.5B | Apache-2.0 | 是 | +| **LTX Video 2** | 19B | Apache-2.0 | 是 | +| **Lumina2** | 2B | Apache-2.0 | 是 | +| **Mage-Flow** | 4B | MIT | 是 | +| **OmniGen** | 3.8B | MIT | 是 | +| **PixArt Sigma** | 0.6B-0.9B | OpenRAIL++ | 是（受限） | +| **Qwen Image** | 20B | Apache-2.0 | 是 | +| **Sana** | 0.6B-4.8B | Apache-2.0 | 是 | +| **Sana Video** | 2B | Apache-2.0 | 是 | +| **SD 1.x/2.x (Legacy)** | 0.9B | OpenRAIL++ | 是（受限） | +| **Stable Diffusion 3** | 2B-8B | Stability AI Community License | 是（年收入低于 100 万美元） | +| **Stable Diffusion XL** | 3.5B | CreativeML OpenRAIL-M | 是（受限） | +| **Stable Cascade (Stage C)** | 1B, 3.6B prior | SimpleTuner 中未指定 | Abandonware | +| **Wan Video** | 1.3B-14B | Apache-2.0 | 是 | +| **Wan S2V** | 14B | Apache-2.0 | 是 | +| **Z-Image** | 6B | Apache-2.0 | 是 | +| **Z-Image Omni** | 6B | Apache-2.0 | 是 | +| **ZLab I1** | 3B | MIT | 是 | + +*许可证值在可用时来自 SimpleTuner 模型 helper，并为此前未指定的条目参考上游模型卡/许可证。`SimpleTuner 中未指定` 表示 helper 未命名许可证，且此处未汇总上游条款；使用前请查看上游模型卡。* ### 高级训练技术 From 66da4d4fa7f9b25308b4d6965139cb3ef6a60699 Mon Sep 17 00:00:00 2001 From: bghira Date: Mon, 3 Aug 2026 13:20:58 -0600 Subject: [PATCH 2/2] Refactor quickstart compatibility tables --- documentation/QUICKSTART.es.md | 326 +++++++++++++++++++++++++---- documentation/QUICKSTART.hi.md | 328 ++++++++++++++++++++++++----- documentation/QUICKSTART.ja.md | 330 +++++++++++++++++++++++++----- documentation/QUICKSTART.md | 326 +++++++++++++++++++++++++---- documentation/QUICKSTART.pt-BR.md | 326 +++++++++++++++++++++++++---- documentation/QUICKSTART.zh.md | 326 +++++++++++++++++++++++++---- 6 files changed, 1689 insertions(+), 273 deletions(-) diff --git a/documentation/QUICKSTART.es.md b/documentation/QUICKSTART.es.md index 5083594a6..d5e2fe470 100644 --- a/documentation/QUICKSTART.es.md +++ b/documentation/QUICKSTART.es.md @@ -2,55 +2,291 @@ **Nota**: Para configuraciones más avanzadas, consulta el [tutorial](TUTORIAL.md) y la [referencia de opciones](OPTIONS.md). +## Guías de inicio rápido por modelo + +| Modelo | Parámetros | Guía | +| --- | --- | --- | +| ACE-Step | 3.5B | [ACE_STEP.es.md](quickstart/ACE_STEP.es.md) | +| Anima | Not specified | Sin guía dedicada | +| Auraflow | 6B | [AURAFLOW.es.md](quickstart/AURAFLOW.es.md) | +| Boogu-Image | Not specified | [BOOGU_IMAGE.es.md](quickstart/BOOGU_IMAGE.es.md) | +| Chroma 1 | 8.9B | [CHROMA.es.md](quickstart/CHROMA.es.md) | +| Cosmos2 | 2B-14B | [COSMOS2IMAGE.es.md](quickstart/COSMOS2IMAGE.es.md) | +| Cosmos3 | 16B-65B | [COSMOS3.es.md](quickstart/COSMOS3.es.md) | +| DeepFloyd IF | 0.4B-4.3B stages | Sin guía dedicada | +| ERNIE-Image | Not specified | [ERNIE.es.md](quickstart/ERNIE.es.md) | +| Flux.1 | 8B-12B | [FLUX.es.md](quickstart/FLUX.es.md)
[FLUX_KONTEXT.es.md](quickstart/FLUX_KONTEXT.es.md) | +| Flux.2 | 4B-32B | [FLUX2.es.md](quickstart/FLUX2.es.md) | +| HeartMuLa | 3B | [HEARTMULA.es.md](quickstart/HEARTMULA.es.md) | +| HiDream | 17B (8.5B MoE) | [HIDREAM.es.md](quickstart/HIDREAM.es.md) | +| Hunyuan Video | 8.3B | [HUNYUANVIDEO.es.md](quickstart/HUNYUANVIDEO.es.md) | +| Ideogram 4 | 9B | [IDEOGRAM4.es.md](quickstart/IDEOGRAM4.es.md) | +| Kandinsky 5.0 Image | 6B (lite) | [KANDINSKY5_IMAGE.es.md](quickstart/KANDINSKY5_IMAGE.es.md) | +| Kandinsky 5.0 Video | 2B lite, 19B pro | [KANDINSKY5_VIDEO.es.md](quickstart/KANDINSKY5_VIDEO.es.md) | +| Kwai Kolors | 2.7B | [KOLORS.es.md](quickstart/KOLORS.es.md) | +| Krea2 | Not specified | [KREA2.es.md](quickstart/KREA2.es.md) | +| LongCat Image | 6B | [LONGCAT_IMAGE.es.md](quickstart/LONGCAT_IMAGE.es.md)
[LONGCAT_EDIT.es.md](quickstart/LONGCAT_EDIT.es.md) | +| LongCat Video | 13.6B | [LONGCAT_VIDEO.es.md](quickstart/LONGCAT_VIDEO.es.md)
[LONGCAT_VIDEO_EDIT.es.md](quickstart/LONGCAT_VIDEO_EDIT.es.md) | +| LTX Video | ~2.5B | [LTXVIDEO.es.md](quickstart/LTXVIDEO.es.md) | +| LTX Video 2 | 19B | [LTXVIDEO2.es.md](quickstart/LTXVIDEO2.es.md) | +| Lumina2 | 2B | [LUMINA2.es.md](quickstart/LUMINA2.es.md) | +| Mage-Flow | 4B | [MAGEFLOW.es.md](quickstart/MAGEFLOW.es.md) | +| OmniGen | 3.8B | [OMNIGEN.es.md](quickstart/OMNIGEN.es.md) | +| PixArt Sigma | 0.6B-0.9B | [SIGMA.es.md](quickstart/SIGMA.es.md) | +| Qwen Image | 20B | [QWEN_IMAGE.es.md](quickstart/QWEN_IMAGE.es.md)
[QWEN_EDIT.es.md](quickstart/QWEN_EDIT.es.md) | +| Sana | 0.6B-4.8B | [SANA.es.md](quickstart/SANA.es.md) | +| Sana Video | 2B | [SANAVIDEO.es.md](quickstart/SANAVIDEO.es.md) | +| SD 1.x/2.x (Legacy) | 0.9B | Sin guía dedicada | +| Stable Diffusion 3 | 2B-8B | [SD3.es.md](quickstart/SD3.es.md) | +| Stable Diffusion XL | 3.5B | [SDXL.es.md](quickstart/SDXL.es.md) | +| Stable Cascade (Stage C) | 1B, 3.6B prior | [STABLE_CASCADE_C.es.md](quickstart/STABLE_CASCADE_C.es.md) | +| Wan Video | 1.3B-14B | [WAN.es.md](quickstart/WAN.es.md) | +| Wan S2V | 14B | [WAN_S2V.es.md](quickstart/WAN_S2V.es.md) | +| Z-Image | 6B | [ZIMAGE.es.md](quickstart/ZIMAGE.es.md) | +| Z-Image Omni | 6B | [ZIMAGE.es.md](quickstart/ZIMAGE.es.md) | +| ZLab I1 | 3B | [ZLAB_i1.es.md](quickstart/ZLAB_i1.es.md) | + ## Compatibilidad de funciones -Para la matriz de funciones completa y más precisa, consulta el [README principal](https://github.com/bghira/SimpleTuner#model-architecture-support). +La matriz completa de compatibilidad se divide por área de función para que cada tabla sea legible. -## Guías de inicio rápido por modelo +
+Soporte de entrenamiento + +| Modelo | PEFT LoRA | LyCORIS | Rango completo | ControlNet | Ref Inputs | +| --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓* | ✗ | ✗ | +| Anima | ✓ | ✓ | ✓* | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓* | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| Chroma 1 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓* | ✗ | audio opt | +| DeepFloyd IF | ✓ | ✓ | ✓ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Flux.1 | ✓ | ✓ | ✓* | ✓ | ✓ opt (Kontext) | +| Flux.2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| HeartMuLa | ✓ | ✓ | ✓* | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓* | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Ideogram 4 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓* | ✗ | ✓ I2I | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Kwai Kolors | ✓ | ✓ | ✓ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| LongCat Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| LongCat Video | ✓ | ✓ | ✓* | ✗ | ✓ opt/edit | +| LTX Video | ✓ | ✓ | ✓ | ✗ | ✓ I2V | +| LTX Video 2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| OmniGen | ✓ | ✓ | ✓ | ✗ | ✗ | +| PixArt Sigma | ✗ | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| Sana | ✗ | ✓ | ✓ | ✗ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓* | ✓ | ✗ | +| Stable Diffusion XL | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓* | ✗ | ✗ | +| Wan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V/VACE | +| Wan S2V | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image Omni | ✓ | ✓ | ✓* | ✗ | ✓ opt (Edit) | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | ✗ | + +
+ +
+Compatibilidad de precisión + +| Modelo | Cuantización | Precisión mixta | +| --- | --- | --- | +| ACE-Step | int8 optional | bf16 | +| Anima | not specified | bf16 | +| Auraflow | int8/fp8/nf4 optional | bf16 | +| Boogu-Image | fp8 optional | bf16 | +| Chroma 1 | int8/fp8/nf4 optional | bf16 | +| Cosmos2 | int8 optional | bf16 | +| Cosmos3 | no_change first; int8 optional | bf16 | +| DeepFloyd IF | not recommended | bf16 | +| ERNIE-Image | int8 optional | bf16 | +| Flux.1 | int8/fp8/nf4 optional | bf16 | +| Flux.2 | int8/fp8/nf4 optional | bf16 | +| HeartMuLa | int8 optional | bf16 | +| HiDream | int8/fp8/nf4 optional | bf16 | +| Hunyuan Video | int8 optional | bf16 | +| Ideogram 4 | fp8 default, nf4 optional | bf16 | +| Kandinsky 5.0 Image | int8 optional | bf16 | +| Kandinsky 5.0 Video | int8 optional | bf16 | +| Kwai Kolors | not recommended | bf16 | +| Krea2 | int8 optional | bf16 | +| LongCat Image | int8/fp8 optional | bf16 | +| LongCat Video | int8/fp8 optional | bf16 | +| LTX Video | int8/fp8 optional | bf16 | +| LTX Video 2 | int8/fp8 optional | bf16 | +| Lumina2 | int8 optional | bf16 | +| Mage-Flow | fp8 optional | bf16 | +| OmniGen | int8/fp8 optional | bf16 | +| PixArt Sigma | int8 optional | bf16 | +| Qwen Image | required (int8/nf4) | bf16 | +| Sana | int8 optional | bf16 | +| Sana Video | not recommended for full | bf16 | +| SD 1.x/2.x (Legacy) | int8/nf4 optional | bf16 | +| Stable Diffusion 3 | int8/fp8/nf4 optional | bf16 | +| Stable Diffusion XL | int8/nf4 optional | bf16 | +| Stable Cascade (Stage C) | not supported | fp32 required | +| Wan Video | int8 optional | bf16 | +| Wan S2V | int8 optional | bf16 | +| Z-Image | int8 optional | bf16 | +| Z-Image Omni | int8 optional | bf16 | +| ZLab I1 | int8 optional | bf16 | + +
+ +
+Granularidad de checkpointing + +| Modelo | Checkpointing de gradiente | Intervalo | Segment stride | Offload de atención | +| --- | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓ | ✗ | +| Anima | ✓ | ✗ | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓ | ✗ | +| Chroma 1 | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓ | ✗ | +| DeepFloyd IF | ✓ | ✗ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓ | ✗ | +| Flux.1 | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | ✓ | ✗ | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | ✓ | ✓ | ✓ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | ✓ | ✗ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | ✓ | ✓ | ✓ | ✓ | +| LTX Video | ✓ | ✓ | ✓ | ✗ | +| LTX Video 2 | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓ | ✓ | +| OmniGen | ✓ | ✗ | ✗ | ✗ | +| PixArt Sigma | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓ | ✗ | +| Sana | ✓ | ✓ | ✓ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✗ | ✗ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | ✓ | ✗ | ✗ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓ | ✗ | +| Wan Video | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | ✓ | ✓ | ✓ | ✗ | +| Z-Image | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | ✓ | ✗ | ✗ | ✗ | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | + +
+ +
+Flujo, destilación y alineación + +| Modelo | Predicción | Flow Shift | TwinFlow | Self-Flow | LayerSync | Sliders | +| --- | --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Anima | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Auraflow | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Boogu-Image | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Chroma 1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | sample | ✗ | ✗ | ✓ | ✓ | ✓ | +| Cosmos3 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| DeepFloyd IF | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| ERNIE-Image | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Flux.1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | autoregressive next-token | ✗ | ✗ | ✗ | ✗ | ✗ | +| HiDream | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Hunyuan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Kandinsky 5.0 Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Krea2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video 2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Mage-Flow | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| OmniGen | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| PixArt Sigma | epsilon | ✗ | ✗ | ✓ | ✓ | ✓ | +| Qwen Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| SD 1.x/2.x (Legacy) | epsilon / v-pred | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Diffusion 3 | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Cascade (Stage C) | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Wan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Z-Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| ZLab I1 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | + +
+ +
+Encoders de texto y tipos de VAE + +| Modelo | Encoders de texto | Parámetros de encoder | VAE | +| --- | --- | --- | --- | +| ACE-Step | UMT5 Encoder | 0.6B | Music DCAE | +| Anima | Qwen3 0.6B | 0.6B | Qwen Image VAE | +| Auraflow | Pile T5 | not specified | AutoencoderKL | +| Boogu-Image | Qwen3-VL | not specified | AutoencoderKL | +| Chroma 1 | T5 XXL v1.1 | 11B | AutoencoderKL | +| Cosmos2 | T5 11B | 11B | Wan VAE | +| Cosmos3 | Cosmos3 reasoner | not specified | Wan/Cosmos VAE | +| DeepFloyd IF | T5 XXL v1.1 | 11B | None | +| ERNIE-Image | ERNIE text encoder | not specified | Flux.2 VAE | +| Flux.1 | CLIP-L/14 + T5 XXL v1.1 | 123M + 11B | AutoencoderKL | +| Flux.2 | Mistral-Small-3.1-24B | 24B | Flux.2 VAE | +| HeartMuLa | None | N/A | HeartCodec tokens | +| HiDream | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 + Llama | 123M + 694M + 11B + not specified | AutoencoderKL | +| Hunyuan Video | Hunyuan LLM | not specified | Hunyuan Video 3D VAE | +| Ideogram 4 | Qwen3-VL-8B-Instruct | 8B | Ideogram AutoEncoder | +| Kandinsky 5.0 Image | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Flux VAE (AutoencoderKL) | +| Kandinsky 5.0 Video | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Hunyuan Video VAE | +| Kwai Kolors | ChatGLM-6B | 6B | AutoencoderKL | +| Krea2 | Qwen3VL | not specified | Qwen Image VAE | +| LongCat Image | Qwen2.5-VL | 7B | AutoencoderKL | +| LongCat Video | Qwen2.5-VL | 7B | Wan VAE | +| LTX Video | T5 XXL v1.1 | 11B | LTX Video VAE | +| LTX Video 2 | Gemma3 | not specified | LTX Video 2 VAE | +| Lumina2 | Gemma2 | 2B | AutoencoderKL | +| Mage-Flow | Qwen3-VL | not specified | Mage-VAE | +| OmniGen | Integrated OmniGen encoder | not specified | AutoencoderKL | +| PixArt Sigma | T5 XXL v1.1 | 11B | AutoencoderKL | +| Qwen Image | Qwen2.5-VL | 7B | Qwen Image VAE | +| Sana | Gemma2 2B-IT | 2B | Sana AutoencoderDC | +| Sana Video | Gemma 2 | 2B | Wan VAE | +| SD 1.x/2.x (Legacy) | CLIP-L/14 | 123M | AutoencoderKL | +| Stable Diffusion 3 | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 | 123M + 694M + 11B | AutoencoderKL | +| Stable Diffusion XL | CLIP-L/14 + CLIP-G/14 | 123M + 694M | AutoencoderKL | +| Stable Cascade (Stage C) | CLIP-ViT-bigG-14 | 694M | Stable Cascade Stage C VAE | +| Wan Video | UMT5 | not specified | Wan VAE | +| Wan S2V | UMT5 | not specified | Wan VAE | +| Z-Image | Qwen3 4B | 4B | AutoencoderKL | +| Z-Image Omni | Qwen3 4B | 4B | AutoencoderKL | +| ZLab I1 | T5Gemma 2B | 2B | AutoencoderKL | + +
-| Modelo | Parámetros | PEFT LoRA | Lycoris | Rango completo | Cuantización | Precisión mixta | Checkpointing de gradiente | Flow Shift | TwinFlow | Self-Flow | LayerSync | Ref Inputs | ControlNet | Sliders† | Guía | -| --- | --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | --- | -| PixArt Sigma | 0.6B–0.9B | ✗ | ✓ | ✓ | int8 opcional | bf16 | ✓ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ | ✓ | [SIGMA.md](quickstart/SIGMA.md) | -| NVLabs Sana | 1.6B–4.8B | ✗ | ✓ | ✓ | int8 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [SANA.md](quickstart/SANA.md) | -| Kwai Kolors | 2.7B | ✓ | ✓ | ✓ | no recomendado | bf16 | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [KOLORS.md](quickstart/KOLORS.md) | -| Stable Diffusion 3 | 2B–8B | ✓ | ✓ | ✓ | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [SD3.md](quickstart/SD3.md) | -| Flux.1 | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [FLUX.md](quickstart/FLUX.md) | -| Flux.2 | 32B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [FLUX2.md](quickstart/FLUX2.md) | -| Flux Kontext | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✓ | ✓ | [FLUX_KONTEXT.md](quickstart/FLUX_KONTEXT.md) | -| Z-Image Turbo | 6B | ✓ | ✗ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZIMAGE.md](quickstart/ZIMAGE.md) | -| Krea2 | - | ✓ | ✗ | ✓* | int8 opcional | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✓ opt | ✗ | ✓ | [KREA2.md](quickstart/KREA2.es.md) | -| Boogu-Image 0.1 | - | ✓ | ✓ | ✓* | fp8 opcional | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ edit | ✗ | ✓ | [BOOGU_IMAGE.md](quickstart/BOOGU_IMAGE.es.md) | -| zlab i1 | 3B | ✓ | ✓ | ✓ | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZLAB_i1.md](quickstart/ZLAB_i1.es.md) | -| Ideogram 4 | 9B | ✓ | ✓ | ✓* | fp8 predeterminado, nf4 opcional | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [IDEOGRAM4.md](quickstart/IDEOGRAM4.es.md) | -| ACE-Step | 3.5B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✓ | [ACE_STEP.md](quickstart/ACE_STEP.md) | -| Chroma 1 | 8.9B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [CHROMA.md](quickstart/CHROMA.md) | -| Auraflow | 6B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [AURAFLOW.md](quickstart/AURAFLOW.md) | -| HiDream I1 | 17B (8.5B MoE) | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [HIDREAM.md](quickstart/HIDREAM.md) | -| OmniGen | 3.8B | ✓ | ✓ | ✓ | int8/fp8 opcional | bf16 | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | [OMNIGEN.md](quickstart/OMNIGEN.md) | -| Stable Diffusion XL | 2.6B | ✓ | ✓ | ✓ | no recomendado | bf16 | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | [SDXL.md](quickstart/SDXL.md) | -| Lumina2 | 2B | ✓ | ✓ | ✓ | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | [LUMINA2.md](quickstart/LUMINA2.md) | -| Cosmos2 | 2B | ✓ | ✓ | ✓ | no recomendado | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [COSMOS2IMAGE.md](quickstart/COSMOS2IMAGE.md) | -| Cosmos3 | 16B-65B | ✓ | ✓ | ✓* | no_change primero | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | audio opt | ✗ | ✓ | [COSMOS3.md](quickstart/COSMOS3.es.md) | -| LTX Video | ~2.5B | ✓ | ✓ | ✓ | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [LTXVIDEO.md](quickstart/LTXVIDEO.md) | -| LTX Video 2 | 19B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LTXVIDEO2.md](quickstart/LTXVIDEO2.md) | -| Hunyuan Video 1.5 | 8.3B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [HUNYUANVIDEO.md](quickstart/HUNYUANVIDEO.md) | -| Wan 2.x | 1.3B–14B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN.md](quickstart/WAN.md) | -| Wan 2.2 S2V | 14B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN_S2V.md](quickstart/WAN_S2V.md) | -| Qwen Image | 20B | ✓ | ✓ | ✓* | **requerido** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [QWEN_IMAGE.md](quickstart/QWEN_IMAGE.md) | -| Qwen Image Edit | 20B | ✓ | ✓ | ✓* | **requerido** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [QWEN_EDIT.md](quickstart/QWEN_EDIT.md) | -| Stable Cascade (C) | 1B, prior 3.6B | ✓ | ✓ | ✓* | no soportado | fp32 (requerido) | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [STABLE_CASCADE_C.md](quickstart/STABLE_CASCADE_C.md) | -| Kandinsky 5.0 Image | 6B (lite) | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ I2I | ✗ | ✓ | [KANDINSKY5_IMAGE.md](quickstart/KANDINSKY5_IMAGE.md) | -| Kandinsky 5.0 Video | 2B (lite), 19B (pro) | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [KANDINSKY5_VIDEO.md](quickstart/KANDINSKY5_VIDEO.md) | -| LongCat-Video | 13.6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LONGCAT_VIDEO.md](quickstart/LONGCAT_VIDEO.md) | -| LongCat-Video Edit | 13.6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_VIDEO_EDIT.md](quickstart/LONGCAT_VIDEO_EDIT.md) | -| LongCat-Image | 6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [LONGCAT_IMAGE.md](quickstart/LONGCAT_IMAGE.md) | -| LongCat-Image Edit | 6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_EDIT.md](quickstart/LONGCAT_EDIT.md) | - -*✓ = soportado, ✓* = requiere DeepSpeed/FSDP2 para rango completo, ✗ = no soportado, `✓+` indica que se recomienda checkpointing por presión de VRAM. Ref Inputs marca rutas existentes de condicionamiento por referencia/edición/I2V; `opt` significa opcional y `req` significa requerido por el flavour de edición/I2V. TwinFlow ✓ significa soporte nativo cuando `twinflow_enabled=true` (los modelos de difusión necesitan `diff2flow_enabled+twinflow_allow_diff2flow`). Self-Flow ✓ significa soporte nativo para `crepa_enabled=true` con `crepa_feature_source=self_flow`, `use_ema=true` y `crepa_teacher_block_index` configurado. LayerSync ✓ significa que el backbone expone estados ocultos del transformer para autoalineación; ✗ marca backbones tipo UNet sin ese buffer. †Sliders aplican a LoRA y LyCORIS (incluido LyCORIS de rango completo “full”).* - -> ℹ️ El inicio rápido de Wan incluye presets de las etapas 2.1 y 2.2 y el toggle de time-embedding. Flux Kontext cubre flujos de edición construidos sobre Flux.1. - -> ⚠️ Estos quickstarts son documentos vivos. Espera actualizaciones ocasionales a medida que llegan nuevos modelos o se mejoran recetas de entrenamiento. +*✓ = soportado, ✓* = soportado pero normalmente requiere DeepSpeed/FSDP2 para entrenamiento full-rank, ✗ = no soportado. Ref Inputs marca rutas existentes de condicionamiento por referencia/edición/I2V; `opt` significa opcional y `req` significa requerido por el flavour de edición/I2V.* +*TwinFlow es nativo cuando `twinflow_enabled=true`; los modelos de difusión aún requieren `diff2flow_enabled=true` y `twinflow_allow_diff2flow=true`. Self-Flow se refiere al soporte CREPA self-flow. LayerSync marca backbones que exponen estados ocultos para alineación.* ### Rutas rápidas: Z-Image Turbo y Flux Schnell diff --git a/documentation/QUICKSTART.hi.md b/documentation/QUICKSTART.hi.md index 2836acf03..59d570f30 100644 --- a/documentation/QUICKSTART.hi.md +++ b/documentation/QUICKSTART.hi.md @@ -2,55 +2,291 @@ **नोट**: अधिक उन्नत कॉन्फ़िगरेशनों के लिए, [ट्यूटोरियल](TUTORIAL.md) और [options reference](OPTIONS.md) देखें। -## फ़ीचर संगतता +## मॉडल क्विकस्टार्ट गाइड -पूरा और सबसे सटीक फीचर मैट्रिक्स देखने के लिए, [मुख्य README](https://github.com/bghira/SimpleTuner#model-architecture-support) देखें। +| मॉडल | पैरामीटर | गाइड | +| --- | --- | --- | +| ACE-Step | 3.5B | [ACE_STEP.hi.md](quickstart/ACE_STEP.hi.md) | +| Anima | Not specified | Dedicated guide नहीं है | +| Auraflow | 6B | [AURAFLOW.hi.md](quickstart/AURAFLOW.hi.md) | +| Boogu-Image | Not specified | [BOOGU_IMAGE.hi.md](quickstart/BOOGU_IMAGE.hi.md) | +| Chroma 1 | 8.9B | [CHROMA.hi.md](quickstart/CHROMA.hi.md) | +| Cosmos2 | 2B-14B | [COSMOS2IMAGE.hi.md](quickstart/COSMOS2IMAGE.hi.md) | +| Cosmos3 | 16B-65B | [COSMOS3.hi.md](quickstart/COSMOS3.hi.md) | +| DeepFloyd IF | 0.4B-4.3B stages | Dedicated guide नहीं है | +| ERNIE-Image | Not specified | [ERNIE.hi.md](quickstart/ERNIE.hi.md) | +| Flux.1 | 8B-12B | [FLUX.hi.md](quickstart/FLUX.hi.md)
[FLUX_KONTEXT.hi.md](quickstart/FLUX_KONTEXT.hi.md) | +| Flux.2 | 4B-32B | [FLUX2.hi.md](quickstart/FLUX2.hi.md) | +| HeartMuLa | 3B | [HEARTMULA.hi.md](quickstart/HEARTMULA.hi.md) | +| HiDream | 17B (8.5B MoE) | [HIDREAM.hi.md](quickstart/HIDREAM.hi.md) | +| Hunyuan Video | 8.3B | [HUNYUANVIDEO.hi.md](quickstart/HUNYUANVIDEO.hi.md) | +| Ideogram 4 | 9B | [IDEOGRAM4.hi.md](quickstart/IDEOGRAM4.hi.md) | +| Kandinsky 5.0 Image | 6B (lite) | [KANDINSKY5_IMAGE.hi.md](quickstart/KANDINSKY5_IMAGE.hi.md) | +| Kandinsky 5.0 Video | 2B lite, 19B pro | [KANDINSKY5_VIDEO.hi.md](quickstart/KANDINSKY5_VIDEO.hi.md) | +| Kwai Kolors | 2.7B | [KOLORS.hi.md](quickstart/KOLORS.hi.md) | +| Krea2 | Not specified | [KREA2.hi.md](quickstart/KREA2.hi.md) | +| LongCat Image | 6B | [LONGCAT_IMAGE.hi.md](quickstart/LONGCAT_IMAGE.hi.md)
[LONGCAT_EDIT.hi.md](quickstart/LONGCAT_EDIT.hi.md) | +| LongCat Video | 13.6B | [LONGCAT_VIDEO.hi.md](quickstart/LONGCAT_VIDEO.hi.md)
[LONGCAT_VIDEO_EDIT.hi.md](quickstart/LONGCAT_VIDEO_EDIT.hi.md) | +| LTX Video | ~2.5B | [LTXVIDEO.hi.md](quickstart/LTXVIDEO.hi.md) | +| LTX Video 2 | 19B | [LTXVIDEO2.hi.md](quickstart/LTXVIDEO2.hi.md) | +| Lumina2 | 2B | [LUMINA2.hi.md](quickstart/LUMINA2.hi.md) | +| Mage-Flow | 4B | [MAGEFLOW.hi.md](quickstart/MAGEFLOW.hi.md) | +| OmniGen | 3.8B | [OMNIGEN.hi.md](quickstart/OMNIGEN.hi.md) | +| PixArt Sigma | 0.6B-0.9B | [SIGMA.hi.md](quickstart/SIGMA.hi.md) | +| Qwen Image | 20B | [QWEN_IMAGE.hi.md](quickstart/QWEN_IMAGE.hi.md)
[QWEN_EDIT.hi.md](quickstart/QWEN_EDIT.hi.md) | +| Sana | 0.6B-4.8B | [SANA.hi.md](quickstart/SANA.hi.md) | +| Sana Video | 2B | [SANAVIDEO.hi.md](quickstart/SANAVIDEO.hi.md) | +| SD 1.x/2.x (Legacy) | 0.9B | Dedicated guide नहीं है | +| Stable Diffusion 3 | 2B-8B | [SD3.hi.md](quickstart/SD3.hi.md) | +| Stable Diffusion XL | 3.5B | [SDXL.hi.md](quickstart/SDXL.hi.md) | +| Stable Cascade (Stage C) | 1B, 3.6B prior | [STABLE_CASCADE_C.hi.md](quickstart/STABLE_CASCADE_C.hi.md) | +| Wan Video | 1.3B-14B | [WAN.hi.md](quickstart/WAN.hi.md) | +| Wan S2V | 14B | [WAN_S2V.hi.md](quickstart/WAN_S2V.hi.md) | +| Z-Image | 6B | [ZIMAGE.hi.md](quickstart/ZIMAGE.hi.md) | +| Z-Image Omni | 6B | [ZIMAGE.hi.md](quickstart/ZIMAGE.hi.md) | +| ZLab I1 | 3B | [ZLAB_i1.hi.md](quickstart/ZLAB_i1.hi.md) | -## मॉडल क्विकस्टार्ट गाइड +## फीचर संगतता + +पूरी compatibility matrix को feature area के अनुसार बांटा गया है ताकि हर table पढ़ने योग्य रहे। + +
+Training support + +| मॉडल | PEFT LoRA | LyCORIS | Full-Rank | ControlNet | Ref Inputs | +| --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓* | ✗ | ✗ | +| Anima | ✓ | ✓ | ✓* | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓* | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| Chroma 1 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓* | ✗ | audio opt | +| DeepFloyd IF | ✓ | ✓ | ✓ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Flux.1 | ✓ | ✓ | ✓* | ✓ | ✓ opt (Kontext) | +| Flux.2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| HeartMuLa | ✓ | ✓ | ✓* | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓* | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Ideogram 4 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓* | ✗ | ✓ I2I | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Kwai Kolors | ✓ | ✓ | ✓ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| LongCat Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| LongCat Video | ✓ | ✓ | ✓* | ✗ | ✓ opt/edit | +| LTX Video | ✓ | ✓ | ✓ | ✗ | ✓ I2V | +| LTX Video 2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| OmniGen | ✓ | ✓ | ✓ | ✗ | ✗ | +| PixArt Sigma | ✗ | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| Sana | ✗ | ✓ | ✓ | ✗ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓* | ✓ | ✗ | +| Stable Diffusion XL | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓* | ✗ | ✗ | +| Wan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V/VACE | +| Wan S2V | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image Omni | ✓ | ✓ | ✓* | ✗ | ✓ opt (Edit) | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | ✗ | + +
+ +
+Precision level support + +| मॉडल | Quantization | Mixed Precision | +| --- | --- | --- | +| ACE-Step | int8 optional | bf16 | +| Anima | not specified | bf16 | +| Auraflow | int8/fp8/nf4 optional | bf16 | +| Boogu-Image | fp8 optional | bf16 | +| Chroma 1 | int8/fp8/nf4 optional | bf16 | +| Cosmos2 | int8 optional | bf16 | +| Cosmos3 | no_change first; int8 optional | bf16 | +| DeepFloyd IF | not recommended | bf16 | +| ERNIE-Image | int8 optional | bf16 | +| Flux.1 | int8/fp8/nf4 optional | bf16 | +| Flux.2 | int8/fp8/nf4 optional | bf16 | +| HeartMuLa | int8 optional | bf16 | +| HiDream | int8/fp8/nf4 optional | bf16 | +| Hunyuan Video | int8 optional | bf16 | +| Ideogram 4 | fp8 default, nf4 optional | bf16 | +| Kandinsky 5.0 Image | int8 optional | bf16 | +| Kandinsky 5.0 Video | int8 optional | bf16 | +| Kwai Kolors | not recommended | bf16 | +| Krea2 | int8 optional | bf16 | +| LongCat Image | int8/fp8 optional | bf16 | +| LongCat Video | int8/fp8 optional | bf16 | +| LTX Video | int8/fp8 optional | bf16 | +| LTX Video 2 | int8/fp8 optional | bf16 | +| Lumina2 | int8 optional | bf16 | +| Mage-Flow | fp8 optional | bf16 | +| OmniGen | int8/fp8 optional | bf16 | +| PixArt Sigma | int8 optional | bf16 | +| Qwen Image | required (int8/nf4) | bf16 | +| Sana | int8 optional | bf16 | +| Sana Video | not recommended for full | bf16 | +| SD 1.x/2.x (Legacy) | int8/nf4 optional | bf16 | +| Stable Diffusion 3 | int8/fp8/nf4 optional | bf16 | +| Stable Diffusion XL | int8/nf4 optional | bf16 | +| Stable Cascade (Stage C) | not supported | fp32 required | +| Wan Video | int8 optional | bf16 | +| Wan S2V | int8 optional | bf16 | +| Z-Image | int8 optional | bf16 | +| Z-Image Omni | int8 optional | bf16 | +| ZLab I1 | int8 optional | bf16 | + +
+ +
+Checkpointing granularity + +| मॉडल | Gradient Checkpoint | Interval | Segment Stride | Attention Offload | +| --- | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓ | ✗ | +| Anima | ✓ | ✗ | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓ | ✗ | +| Chroma 1 | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓ | ✗ | +| DeepFloyd IF | ✓ | ✗ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓ | ✗ | +| Flux.1 | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | ✓ | ✗ | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | ✓ | ✓ | ✓ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | ✓ | ✗ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | ✓ | ✓ | ✓ | ✓ | +| LTX Video | ✓ | ✓ | ✓ | ✗ | +| LTX Video 2 | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓ | ✓ | +| OmniGen | ✓ | ✗ | ✗ | ✗ | +| PixArt Sigma | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓ | ✗ | +| Sana | ✓ | ✓ | ✓ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✗ | ✗ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | ✓ | ✗ | ✗ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓ | ✗ | +| Wan Video | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | ✓ | ✓ | ✓ | ✗ | +| Z-Image | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | ✓ | ✗ | ✗ | ✗ | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | + +
+ +
+Flow, distillation, and alignment + +| मॉडल | Prediction | Flow Shift | TwinFlow | Self-Flow | LayerSync | Sliders | +| --- | --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Anima | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Auraflow | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Boogu-Image | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Chroma 1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | sample | ✗ | ✗ | ✓ | ✓ | ✓ | +| Cosmos3 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| DeepFloyd IF | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| ERNIE-Image | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Flux.1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | autoregressive next-token | ✗ | ✗ | ✗ | ✗ | ✗ | +| HiDream | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Hunyuan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Kandinsky 5.0 Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Krea2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video 2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Mage-Flow | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| OmniGen | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| PixArt Sigma | epsilon | ✗ | ✗ | ✓ | ✓ | ✓ | +| Qwen Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| SD 1.x/2.x (Legacy) | epsilon / v-pred | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Diffusion 3 | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Cascade (Stage C) | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Wan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Z-Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| ZLab I1 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | + +
+ +
+Text encoders and VAE types + +| मॉडल | Text Encoders | Text Encoder Params | VAE | +| --- | --- | --- | --- | +| ACE-Step | UMT5 Encoder | 0.6B | Music DCAE | +| Anima | Qwen3 0.6B | 0.6B | Qwen Image VAE | +| Auraflow | Pile T5 | not specified | AutoencoderKL | +| Boogu-Image | Qwen3-VL | not specified | AutoencoderKL | +| Chroma 1 | T5 XXL v1.1 | 11B | AutoencoderKL | +| Cosmos2 | T5 11B | 11B | Wan VAE | +| Cosmos3 | Cosmos3 reasoner | not specified | Wan/Cosmos VAE | +| DeepFloyd IF | T5 XXL v1.1 | 11B | None | +| ERNIE-Image | ERNIE text encoder | not specified | Flux.2 VAE | +| Flux.1 | CLIP-L/14 + T5 XXL v1.1 | 123M + 11B | AutoencoderKL | +| Flux.2 | Mistral-Small-3.1-24B | 24B | Flux.2 VAE | +| HeartMuLa | None | N/A | HeartCodec tokens | +| HiDream | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 + Llama | 123M + 694M + 11B + not specified | AutoencoderKL | +| Hunyuan Video | Hunyuan LLM | not specified | Hunyuan Video 3D VAE | +| Ideogram 4 | Qwen3-VL-8B-Instruct | 8B | Ideogram AutoEncoder | +| Kandinsky 5.0 Image | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Flux VAE (AutoencoderKL) | +| Kandinsky 5.0 Video | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Hunyuan Video VAE | +| Kwai Kolors | ChatGLM-6B | 6B | AutoencoderKL | +| Krea2 | Qwen3VL | not specified | Qwen Image VAE | +| LongCat Image | Qwen2.5-VL | 7B | AutoencoderKL | +| LongCat Video | Qwen2.5-VL | 7B | Wan VAE | +| LTX Video | T5 XXL v1.1 | 11B | LTX Video VAE | +| LTX Video 2 | Gemma3 | not specified | LTX Video 2 VAE | +| Lumina2 | Gemma2 | 2B | AutoencoderKL | +| Mage-Flow | Qwen3-VL | not specified | Mage-VAE | +| OmniGen | Integrated OmniGen encoder | not specified | AutoencoderKL | +| PixArt Sigma | T5 XXL v1.1 | 11B | AutoencoderKL | +| Qwen Image | Qwen2.5-VL | 7B | Qwen Image VAE | +| Sana | Gemma2 2B-IT | 2B | Sana AutoencoderDC | +| Sana Video | Gemma 2 | 2B | Wan VAE | +| SD 1.x/2.x (Legacy) | CLIP-L/14 | 123M | AutoencoderKL | +| Stable Diffusion 3 | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 | 123M + 694M + 11B | AutoencoderKL | +| Stable Diffusion XL | CLIP-L/14 + CLIP-G/14 | 123M + 694M | AutoencoderKL | +| Stable Cascade (Stage C) | CLIP-ViT-bigG-14 | 694M | Stable Cascade Stage C VAE | +| Wan Video | UMT5 | not specified | Wan VAE | +| Wan S2V | UMT5 | not specified | Wan VAE | +| Z-Image | Qwen3 4B | 4B | AutoencoderKL | +| Z-Image Omni | Qwen3 4B | 4B | AutoencoderKL | +| ZLab I1 | T5Gemma 2B | 2B | AutoencoderKL | + +
-| मॉडल | पैरामीटर | PEFT LoRA | Lycoris | फुल-रैंक | क्वांटाइज़ेशन | मिक्स्ड प्रिसिजन | ग्रैड चेकपॉइंट | फ्लो शिफ्ट | TwinFlow | Self-Flow | LayerSync | Ref Inputs | ControlNet | Sliders† | गाइड | -| --- | --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | --- | -| PixArt Sigma | 0.6B–0.9B | ✗ | ✓ | ✓ | int8 वैकल्पिक | bf16 | ✓ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ | ✓ | [SIGMA.md](quickstart/SIGMA.md) | -| NVLabs Sana | 1.6B–4.8B | ✗ | ✓ | ✓ | int8 वैकल्पिक | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [SANA.md](quickstart/SANA.md) | -| Kwai Kolors | 2.7B | ✓ | ✓ | ✓ | अनुशंसित नहीं | bf16 | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [KOLORS.md](quickstart/KOLORS.md) | -| Stable Diffusion 3 | 2B–8B | ✓ | ✓ | ✓ | int8/fp8/nf4 वैकल्पिक | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [SD3.md](quickstart/SD3.md) | -| Flux.1 | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 वैकल्पिक | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [FLUX.md](quickstart/FLUX.md) | -| Flux.2 | 32B | ✓ | ✓ | ✓* | int8/fp8/nf4 वैकल्पिक | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [FLUX2.md](quickstart/FLUX2.md) | -| Flux Kontext | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 वैकल्पिक | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✓ | ✓ | [FLUX_KONTEXT.md](quickstart/FLUX_KONTEXT.md) | -| Z-Image Turbo | 6B | ✓ | ✗ | ✓* | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZIMAGE.md](quickstart/ZIMAGE.md) | -| Krea2 | - | ✓ | ✗ | ✓* | int8 वैकल्पिक | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✓ opt | ✗ | ✓ | [KREA2.md](quickstart/KREA2.hi.md) | -| Boogu-Image 0.1 | - | ✓ | ✓ | ✓* | fp8 वैकल्पिक | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ edit | ✗ | ✓ | [BOOGU_IMAGE.md](quickstart/BOOGU_IMAGE.hi.md) | -| zlab i1 | 3B | ✓ | ✓ | ✓ | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZLAB_i1.md](quickstart/ZLAB_i1.hi.md) | -| Ideogram 4 | 9B | ✓ | ✓ | ✓* | fp8 डिफ़ॉल्ट, nf4 वैकल्पिक | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [IDEOGRAM4.md](quickstart/IDEOGRAM4.hi.md) | -| ACE-Step | 3.5B | ✓ | ✓ | ✓* | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✓ | [ACE_STEP.md](quickstart/ACE_STEP.md) | -| Chroma 1 | 8.9B | ✓ | ✓ | ✓* | int8/fp8/nf4 वैकल्पिक | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [CHROMA.md](quickstart/CHROMA.md) | -| Auraflow | 6B | ✓ | ✓ | ✓* | int8/fp8/nf4 वैकल्पिक | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [AURAFLOW.md](quickstart/AURAFLOW.md) | -| HiDream I1 | 17B (8.5B MoE) | ✓ | ✓ | ✓* | int8/fp8/nf4 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [HIDREAM.md](quickstart/HIDREAM.md) | -| OmniGen | 3.8B | ✓ | ✓ | ✓ | int8/fp8 वैकल्पिक | bf16 | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | [OMNIGEN.md](quickstart/OMNIGEN.md) | -| Stable Diffusion XL | 2.6B | ✓ | ✓ | ✓ | अनुशंसित नहीं | bf16 | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | [SDXL.md](quickstart/SDXL.md) | -| Lumina2 | 2B | ✓ | ✓ | ✓ | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | [LUMINA2.md](quickstart/LUMINA2.md) | -| Cosmos2 | 2B | ✓ | ✓ | ✓ | अनुशंसित नहीं | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [COSMOS2IMAGE.md](quickstart/COSMOS2IMAGE.md) | -| Cosmos3 | 16B-65B | ✓ | ✓ | ✓* | no_change first | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | audio opt | ✗ | ✓ | [COSMOS3.md](quickstart/COSMOS3.hi.md) | -| LTX Video | ~2.5B | ✓ | ✓ | ✓ | int8/fp8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [LTXVIDEO.md](quickstart/LTXVIDEO.md) | -| LTX Video 2 | 19B | ✓ | ✓ | ✓* | int8/fp8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LTXVIDEO2.md](quickstart/LTXVIDEO2.md) | -| Hunyuan Video 1.5 | 8.3B | ✓ | ✓ | ✓* | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [HUNYUANVIDEO.md](quickstart/HUNYUANVIDEO.md) | -| Wan 2.x | 1.3B–14B | ✓ | ✓ | ✓* | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN.md](quickstart/WAN.md) | -| Wan 2.2 S2V | 14B | ✓ | ✓ | ✓* | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN_S2V.md](quickstart/WAN_S2V.md) | -| Qwen Image | 20B | ✓ | ✓ | ✓* | **आवश्यक** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [QWEN_IMAGE.md](quickstart/QWEN_IMAGE.md) | -| Qwen Image Edit | 20B | ✓ | ✓ | ✓* | **आवश्यक** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [QWEN_EDIT.md](quickstart/QWEN_EDIT.md) | -| Stable Cascade (C) | 1B, 3.6B prior | ✓ | ✓ | ✓* | समर्थित नहीं | fp32 (आवश्यक) | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [STABLE_CASCADE_C.md](quickstart/STABLE_CASCADE_C.md) | -| Kandinsky 5.0 Image | 6B (lite) | ✓ | ✓ | ✓* | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ I2I | ✗ | ✓ | [KANDINSKY5_IMAGE.md](quickstart/KANDINSKY5_IMAGE.md) | -| Kandinsky 5.0 Video | 2B (lite), 19B (pro) | ✓ | ✓ | ✓* | int8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [KANDINSKY5_VIDEO.md](quickstart/KANDINSKY5_VIDEO.md) | -| LongCat-Video | 13.6B | ✓ | ✓ | ✓* | int8/fp8 वैकल्पिक | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LONGCAT_VIDEO.md](quickstart/LONGCAT_VIDEO.md) | -| LongCat-Video Edit | 13.6B | ✓ | ✓ | ✓* | int8/fp8 वैकल्पिक | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_VIDEO_EDIT.md](quickstart/LONGCAT_VIDEO_EDIT.md) | -| LongCat-Image | 6B | ✓ | ✓ | ✓* | int8/fp8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [LONGCAT_IMAGE.md](quickstart/LONGCAT_IMAGE.md) | -| LongCat-Image Edit | 6B | ✓ | ✓ | ✓* | int8/fp8 वैकल्पिक | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_EDIT.md](quickstart/LONGCAT_EDIT.md) | - -*✓ = समर्थित, ✓* = फुल‑रैंक के लिए DeepSpeed/FSDP2 आवश्यक, ✗ = समर्थित नहीं, `✓+` VRAM दबाव के कारण checkpointing की सिफ़ारिश को दर्शाता है। Ref Inputs सिर्फ मौजूदा reference/edit/I2V conditioning paths को दिखाता है; `opt` वैकल्पिक है और `req` edit/I2V flavour के लिए आवश्यक है। TwinFlow ✓ का अर्थ है `twinflow_enabled=true` होने पर native support (diffusion मॉडल्स को `diff2flow_enabled+twinflow_allow_diff2flow` चाहिए)। Self-Flow ✓ का अर्थ है `crepa_enabled=true`, `crepa_feature_source=self_flow`, `use_ema=true`, और `crepa_teacher_block_index` सेट होने पर native support। LayerSync ✓ का अर्थ है कि backbone self‑alignment के लिए transformer hidden states उपलब्ध कराता है; ✗ UNet‑style backbones को दर्शाता है जिनमें वह buffer नहीं होता। †Sliders LoRA और LyCORIS (full‑rank LyCORIS “full” सहित) पर लागू होते हैं।* - -> ℹ️ Wan quickstart में 2.1 + 2.2 stage presets और time‑embedding toggle शामिल है। Flux Kontext में Flux.1 के ऊपर बने editing वर्कफ़्लो शामिल हैं। - -> ⚠️ ये क्विकस्टार्ट living documents हैं। नए मॉडल आने या प्रशिक्षण रेसिपीज़ सुधरने के साथ समय‑समय पर अपडेट की उम्मीद करें। +*✓ = supported, ✓* = supported लेकिन full-rank training के लिए आम तौर पर DeepSpeed/FSDP2 चाहिए, ✗ = not supported. Ref Inputs मौजूदा reference/edit/I2V conditioning paths को दिखाता है; `opt` optional और `req` edit/I2V flavour में required है.* +*TwinFlow native है जब `twinflow_enabled=true`; diffusion models को अभी भी `diff2flow_enabled=true` और `twinflow_allow_diff2flow=true` चाहिए। Self-Flow CREPA self-flow support है। LayerSync alignment के लिए hidden states expose करने वाले backbones को दिखाता है.* ### तेज़ रास्ते: Z-Image Turbo और Flux Schnell diff --git a/documentation/QUICKSTART.ja.md b/documentation/QUICKSTART.ja.md index c3e2fdebe..2ec877d56 100644 --- a/documentation/QUICKSTART.ja.md +++ b/documentation/QUICKSTART.ja.md @@ -2,55 +2,291 @@ **注意**: より高度な設定については、[チュートリアル](TUTORIAL.md)および[オプションリファレンス](OPTIONS.md)を参照してください。 +## モデル別クイックスタートガイド + +| モデル | パラメータ | ガイド | +| --- | --- | --- | +| ACE-Step | 3.5B | [ACE_STEP.ja.md](quickstart/ACE_STEP.ja.md) | +| Anima | Not specified | 専用ガイドなし | +| Auraflow | 6B | [AURAFLOW.ja.md](quickstart/AURAFLOW.ja.md) | +| Boogu-Image | Not specified | [BOOGU_IMAGE.ja.md](quickstart/BOOGU_IMAGE.ja.md) | +| Chroma 1 | 8.9B | [CHROMA.ja.md](quickstart/CHROMA.ja.md) | +| Cosmos2 | 2B-14B | [COSMOS2IMAGE.ja.md](quickstart/COSMOS2IMAGE.ja.md) | +| Cosmos3 | 16B-65B | [COSMOS3.ja.md](quickstart/COSMOS3.ja.md) | +| DeepFloyd IF | 0.4B-4.3B stages | 専用ガイドなし | +| ERNIE-Image | Not specified | [ERNIE.ja.md](quickstart/ERNIE.ja.md) | +| Flux.1 | 8B-12B | [FLUX.ja.md](quickstart/FLUX.ja.md)
[FLUX_KONTEXT.ja.md](quickstart/FLUX_KONTEXT.ja.md) | +| Flux.2 | 4B-32B | [FLUX2.ja.md](quickstart/FLUX2.ja.md) | +| HeartMuLa | 3B | [HEARTMULA.ja.md](quickstart/HEARTMULA.ja.md) | +| HiDream | 17B (8.5B MoE) | [HIDREAM.ja.md](quickstart/HIDREAM.ja.md) | +| Hunyuan Video | 8.3B | [HUNYUANVIDEO.ja.md](quickstart/HUNYUANVIDEO.ja.md) | +| Ideogram 4 | 9B | [IDEOGRAM4.ja.md](quickstart/IDEOGRAM4.ja.md) | +| Kandinsky 5.0 Image | 6B (lite) | [KANDINSKY5_IMAGE.ja.md](quickstart/KANDINSKY5_IMAGE.ja.md) | +| Kandinsky 5.0 Video | 2B lite, 19B pro | [KANDINSKY5_VIDEO.ja.md](quickstart/KANDINSKY5_VIDEO.ja.md) | +| Kwai Kolors | 2.7B | [KOLORS.ja.md](quickstart/KOLORS.ja.md) | +| Krea2 | Not specified | [KREA2.ja.md](quickstart/KREA2.ja.md) | +| LongCat Image | 6B | [LONGCAT_IMAGE.ja.md](quickstart/LONGCAT_IMAGE.ja.md)
[LONGCAT_EDIT.ja.md](quickstart/LONGCAT_EDIT.ja.md) | +| LongCat Video | 13.6B | [LONGCAT_VIDEO.ja.md](quickstart/LONGCAT_VIDEO.ja.md)
[LONGCAT_VIDEO_EDIT.ja.md](quickstart/LONGCAT_VIDEO_EDIT.ja.md) | +| LTX Video | ~2.5B | [LTXVIDEO.ja.md](quickstart/LTXVIDEO.ja.md) | +| LTX Video 2 | 19B | [LTXVIDEO2.ja.md](quickstart/LTXVIDEO2.ja.md) | +| Lumina2 | 2B | [LUMINA2.ja.md](quickstart/LUMINA2.ja.md) | +| Mage-Flow | 4B | [MAGEFLOW.ja.md](quickstart/MAGEFLOW.ja.md) | +| OmniGen | 3.8B | [OMNIGEN.ja.md](quickstart/OMNIGEN.ja.md) | +| PixArt Sigma | 0.6B-0.9B | [SIGMA.ja.md](quickstart/SIGMA.ja.md) | +| Qwen Image | 20B | [QWEN_IMAGE.ja.md](quickstart/QWEN_IMAGE.ja.md)
[QWEN_EDIT.ja.md](quickstart/QWEN_EDIT.ja.md) | +| Sana | 0.6B-4.8B | [SANA.ja.md](quickstart/SANA.ja.md) | +| Sana Video | 2B | [SANAVIDEO.ja.md](quickstart/SANAVIDEO.ja.md) | +| SD 1.x/2.x (Legacy) | 0.9B | 専用ガイドなし | +| Stable Diffusion 3 | 2B-8B | [SD3.ja.md](quickstart/SD3.ja.md) | +| Stable Diffusion XL | 3.5B | [SDXL.ja.md](quickstart/SDXL.ja.md) | +| Stable Cascade (Stage C) | 1B, 3.6B prior | [STABLE_CASCADE_C.ja.md](quickstart/STABLE_CASCADE_C.ja.md) | +| Wan Video | 1.3B-14B | [WAN.ja.md](quickstart/WAN.ja.md) | +| Wan S2V | 14B | [WAN_S2V.ja.md](quickstart/WAN_S2V.ja.md) | +| Z-Image | 6B | [ZIMAGE.ja.md](quickstart/ZIMAGE.ja.md) | +| Z-Image Omni | 6B | [ZIMAGE.ja.md](quickstart/ZIMAGE.ja.md) | +| ZLab I1 | 3B | [ZLAB_i1.ja.md](quickstart/ZLAB_i1.ja.md) | + ## 機能互換性 -完全かつ最も正確な機能マトリックスについては、[メインREADME](https://github.com/bghira/SimpleTuner#model-architecture-support)を参照してください。 - -## モデルクイックスタートガイド - -| モデル | パラメータ数 | PEFT LoRA | Lycoris | Full-Rank | 量子化 | 混合精度 | Grad Checkpoint | Flow Shift | TwinFlow | Self-Flow | LayerSync | Ref Inputs | ControlNet | Sliders† | ガイド | -| --- | --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | --- | -| PixArt Sigma | 0.6B–0.9B | ✗ | ✓ | ✓ | int8 オプション | bf16 | ✓ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ | ✓ | [SIGMA.md](quickstart/SIGMA.md) | -| NVLabs Sana | 1.6B–4.8B | ✗ | ✓ | ✓ | int8 オプション | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [SANA.md](quickstart/SANA.md) | -| Kwai Kolors | 2.7B | ✓ | ✓ | ✓ | 非推奨 | bf16 | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [KOLORS.md](quickstart/KOLORS.md) | -| Stable Diffusion 3 | 2B–8B | ✓ | ✓ | ✓ | int8/fp8/nf4 オプション | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [SD3.md](quickstart/SD3.md) | -| Flux.1 | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 オプション | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [FLUX.md](quickstart/FLUX.md) | -| Flux.2 | 32B | ✓ | ✓ | ✓* | int8/fp8/nf4 オプション | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [FLUX2.md](quickstart/FLUX2.md) | -| Flux Kontext | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 オプション | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✓ | ✓ | [FLUX_KONTEXT.md](quickstart/FLUX_KONTEXT.md) | -| Z-Image Turbo | 6B | ✓ | ✗ | ✓* | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZIMAGE.md](quickstart/ZIMAGE.md) | -| Krea2 | - | ✓ | ✗ | ✓* | int8 オプション | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✓ opt | ✗ | ✓ | [KREA2.md](quickstart/KREA2.ja.md) | -| Boogu-Image 0.1 | - | ✓ | ✓ | ✓* | fp8 オプション | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ edit | ✗ | ✓ | [BOOGU_IMAGE.md](quickstart/BOOGU_IMAGE.ja.md) | -| zlab i1 | 3B | ✓ | ✓ | ✓ | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZLAB_i1.md](quickstart/ZLAB_i1.ja.md) | -| Ideogram 4 | 9B | ✓ | ✓ | ✓* | fp8 デフォルト、nf4 オプション | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [IDEOGRAM4.md](quickstart/IDEOGRAM4.ja.md) | -| ACE-Step | 3.5B | ✓ | ✓ | ✓* | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✓ | [ACE_STEP.md](quickstart/ACE_STEP.md) | -| Chroma 1 | 8.9B | ✓ | ✓ | ✓* | int8/fp8/nf4 オプション | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [CHROMA.md](quickstart/CHROMA.md) | -| Auraflow | 6B | ✓ | ✓ | ✓* | int8/fp8/nf4 オプション | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [AURAFLOW.md](quickstart/AURAFLOW.md) | -| HiDream I1 | 17B (8.5B MoE) | ✓ | ✓ | ✓* | int8/fp8/nf4 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [HIDREAM.md](quickstart/HIDREAM.md) | -| OmniGen | 3.8B | ✓ | ✓ | ✓ | int8/fp8 オプション | bf16 | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | [OMNIGEN.md](quickstart/OMNIGEN.md) | -| Stable Diffusion XL | 2.6B | ✓ | ✓ | ✓ | 非推奨 | bf16 | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | [SDXL.md](quickstart/SDXL.md) | -| Lumina2 | 2B | ✓ | ✓ | ✓ | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | [LUMINA2.md](quickstart/LUMINA2.md) | -| Cosmos2 | 2B | ✓ | ✓ | ✓ | 非推奨 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [COSMOS2IMAGE.md](quickstart/COSMOS2IMAGE.md) | -| Cosmos3 | 16B-65B | ✓ | ✓ | ✓* | no_change first | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | audio opt | ✗ | ✓ | [COSMOS3.md](quickstart/COSMOS3.ja.md) | -| LTX Video | ~2.5B | ✓ | ✓ | ✓ | int8/fp8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [LTXVIDEO.md](quickstart/LTXVIDEO.md) | -| LTX Video 2 | 19B | ✓ | ✓ | ✓* | int8/fp8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LTXVIDEO2.md](quickstart/LTXVIDEO2.md) | -| Hunyuan Video 1.5 | 8.3B | ✓ | ✓ | ✓* | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [HUNYUANVIDEO.md](quickstart/HUNYUANVIDEO.md) | -| Wan 2.x | 1.3B–14B | ✓ | ✓ | ✓* | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN.md](quickstart/WAN.md) | -| Wan 2.2 S2V | 14B | ✓ | ✓ | ✓* | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN_S2V.md](quickstart/WAN_S2V.md) | -| Qwen Image | 20B | ✓ | ✓ | ✓* | **必須** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [QWEN_IMAGE.md](quickstart/QWEN_IMAGE.md) | -| Qwen Image Edit | 20B | ✓ | ✓ | ✓* | **必須** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [QWEN_EDIT.md](quickstart/QWEN_EDIT.md) | -| Stable Cascade (C) | 1B, 3.6B prior | ✓ | ✓ | ✓* | 非対応 | fp32 (必須) | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [STABLE_CASCADE_C.md](quickstart/STABLE_CASCADE_C.md) | -| Kandinsky 5.0 Image | 6B (lite) | ✓ | ✓ | ✓* | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ I2I | ✗ | ✓ | [KANDINSKY5_IMAGE.md](quickstart/KANDINSKY5_IMAGE.md) | -| Kandinsky 5.0 Video | 2B (lite), 19B (pro) | ✓ | ✓ | ✓* | int8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [KANDINSKY5_VIDEO.md](quickstart/KANDINSKY5_VIDEO.md) | -| LongCat-Video | 13.6B | ✓ | ✓ | ✓* | int8/fp8 オプション | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LONGCAT_VIDEO.md](quickstart/LONGCAT_VIDEO.md) | -| LongCat-Video Edit | 13.6B | ✓ | ✓ | ✓* | int8/fp8 オプション | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_VIDEO_EDIT.md](quickstart/LONGCAT_VIDEO_EDIT.md) | -| LongCat-Image | 6B | ✓ | ✓ | ✓* | int8/fp8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [LONGCAT_IMAGE.md](quickstart/LONGCAT_IMAGE.md) | -| LongCat-Image Edit | 6B | ✓ | ✓ | ✓* | int8/fp8 オプション | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_EDIT.md](quickstart/LONGCAT_EDIT.md) | - -*✓ = サポート、✓* = Full-RankにはDeepSpeed/FSDP2が必要、✗ = 非サポート、`✓+`はVRAMプレッシャーによりチェックポイントが推奨されることを示します。Ref Inputs は既存の参照/編集/I2V 条件パスのみを示し、`opt` は任意、`req` は編集/I2V flavour で必須であることを示します。TwinFlow ✓は`twinflow_enabled=true`のときのネイティブサポートを意味します（拡散モデルには`diff2flow_enabled+twinflow_allow_diff2flow`が必要）。Self-Flow ✓は`crepa_enabled=true`、`crepa_feature_source=self_flow`、`use_ema=true`、および`crepa_teacher_block_index`設定時のネイティブサポートを意味します。LayerSync ✓はバックボーンがセルフアライメント用のトランスフォーマー隠れ状態を公開していることを意味し、✗はそのバッファを持たないUNetスタイルのバックボーンを示します。†SlidersはLoRAおよびLyCORIS（Full-Rank LyCORIS "full"を含む）に適用されます。* - -> ℹ️ Wanクイックスタートには2.1 + 2.2ステージプリセットと時間埋め込みトグルが含まれます。Flux Kontextは、Flux.1をベースに構築された編集ワークフローをカバーします。 - -> ⚠️ これらのクイックスタートは生きたドキュメントです。新しいモデルの登場やトレーニングレシピの改善に伴い、時折更新されることがあります。 +完全な互換性マトリクスは、読みやすいように機能領域ごとに分割しています。 + +
+トレーニングサポート + +| モデル | PEFT LoRA | LyCORIS | フルランク | ControlNet | Ref Inputs | +| --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓* | ✗ | ✗ | +| Anima | ✓ | ✓ | ✓* | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓* | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| Chroma 1 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓* | ✗ | audio opt | +| DeepFloyd IF | ✓ | ✓ | ✓ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Flux.1 | ✓ | ✓ | ✓* | ✓ | ✓ opt (Kontext) | +| Flux.2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| HeartMuLa | ✓ | ✓ | ✓* | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓* | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Ideogram 4 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓* | ✗ | ✓ I2I | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Kwai Kolors | ✓ | ✓ | ✓ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| LongCat Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| LongCat Video | ✓ | ✓ | ✓* | ✗ | ✓ opt/edit | +| LTX Video | ✓ | ✓ | ✓ | ✗ | ✓ I2V | +| LTX Video 2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| OmniGen | ✓ | ✓ | ✓ | ✗ | ✗ | +| PixArt Sigma | ✗ | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| Sana | ✗ | ✓ | ✓ | ✗ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓* | ✓ | ✗ | +| Stable Diffusion XL | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓* | ✗ | ✗ | +| Wan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V/VACE | +| Wan S2V | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image Omni | ✓ | ✓ | ✓* | ✗ | ✓ opt (Edit) | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | ✗ | + +
+ +
+精度レベルのサポート + +| モデル | 量子化 | 混合精度 | +| --- | --- | --- | +| ACE-Step | int8 optional | bf16 | +| Anima | not specified | bf16 | +| Auraflow | int8/fp8/nf4 optional | bf16 | +| Boogu-Image | fp8 optional | bf16 | +| Chroma 1 | int8/fp8/nf4 optional | bf16 | +| Cosmos2 | int8 optional | bf16 | +| Cosmos3 | no_change first; int8 optional | bf16 | +| DeepFloyd IF | not recommended | bf16 | +| ERNIE-Image | int8 optional | bf16 | +| Flux.1 | int8/fp8/nf4 optional | bf16 | +| Flux.2 | int8/fp8/nf4 optional | bf16 | +| HeartMuLa | int8 optional | bf16 | +| HiDream | int8/fp8/nf4 optional | bf16 | +| Hunyuan Video | int8 optional | bf16 | +| Ideogram 4 | fp8 default, nf4 optional | bf16 | +| Kandinsky 5.0 Image | int8 optional | bf16 | +| Kandinsky 5.0 Video | int8 optional | bf16 | +| Kwai Kolors | not recommended | bf16 | +| Krea2 | int8 optional | bf16 | +| LongCat Image | int8/fp8 optional | bf16 | +| LongCat Video | int8/fp8 optional | bf16 | +| LTX Video | int8/fp8 optional | bf16 | +| LTX Video 2 | int8/fp8 optional | bf16 | +| Lumina2 | int8 optional | bf16 | +| Mage-Flow | fp8 optional | bf16 | +| OmniGen | int8/fp8 optional | bf16 | +| PixArt Sigma | int8 optional | bf16 | +| Qwen Image | required (int8/nf4) | bf16 | +| Sana | int8 optional | bf16 | +| Sana Video | not recommended for full | bf16 | +| SD 1.x/2.x (Legacy) | int8/nf4 optional | bf16 | +| Stable Diffusion 3 | int8/fp8/nf4 optional | bf16 | +| Stable Diffusion XL | int8/nf4 optional | bf16 | +| Stable Cascade (Stage C) | not supported | fp32 required | +| Wan Video | int8 optional | bf16 | +| Wan S2V | int8 optional | bf16 | +| Z-Image | int8 optional | bf16 | +| Z-Image Omni | int8 optional | bf16 | +| ZLab I1 | int8 optional | bf16 | + +
+ +
+チェックポイント粒度 + +| モデル | Gradient Checkpoint | Interval | Segment Stride | Attention Offload | +| --- | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓ | ✗ | +| Anima | ✓ | ✗ | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓ | ✗ | +| Chroma 1 | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓ | ✗ | +| DeepFloyd IF | ✓ | ✗ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓ | ✗ | +| Flux.1 | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | ✓ | ✗ | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | ✓ | ✓ | ✓ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | ✓ | ✗ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | ✓ | ✓ | ✓ | ✓ | +| LTX Video | ✓ | ✓ | ✓ | ✗ | +| LTX Video 2 | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓ | ✓ | +| OmniGen | ✓ | ✗ | ✗ | ✗ | +| PixArt Sigma | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓ | ✗ | +| Sana | ✓ | ✓ | ✓ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✗ | ✗ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | ✓ | ✗ | ✗ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓ | ✗ | +| Wan Video | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | ✓ | ✓ | ✓ | ✗ | +| Z-Image | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | ✓ | ✗ | ✗ | ✗ | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | + +
+ +
+Flow・蒸留・アラインメント + +| モデル | Prediction | Flow Shift | TwinFlow | Self-Flow | LayerSync | Sliders | +| --- | --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Anima | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Auraflow | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Boogu-Image | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Chroma 1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | sample | ✗ | ✗ | ✓ | ✓ | ✓ | +| Cosmos3 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| DeepFloyd IF | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| ERNIE-Image | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Flux.1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | autoregressive next-token | ✗ | ✗ | ✗ | ✗ | ✗ | +| HiDream | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Hunyuan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Kandinsky 5.0 Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Krea2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video 2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Mage-Flow | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| OmniGen | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| PixArt Sigma | epsilon | ✗ | ✗ | ✓ | ✓ | ✓ | +| Qwen Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| SD 1.x/2.x (Legacy) | epsilon / v-pred | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Diffusion 3 | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Cascade (Stage C) | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Wan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Z-Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| ZLab I1 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | + +
+ +
+テキストエンコーダーと VAE タイプ + +| モデル | Text Encoders | Text Encoder Params | VAE | +| --- | --- | --- | --- | +| ACE-Step | UMT5 Encoder | 0.6B | Music DCAE | +| Anima | Qwen3 0.6B | 0.6B | Qwen Image VAE | +| Auraflow | Pile T5 | not specified | AutoencoderKL | +| Boogu-Image | Qwen3-VL | not specified | AutoencoderKL | +| Chroma 1 | T5 XXL v1.1 | 11B | AutoencoderKL | +| Cosmos2 | T5 11B | 11B | Wan VAE | +| Cosmos3 | Cosmos3 reasoner | not specified | Wan/Cosmos VAE | +| DeepFloyd IF | T5 XXL v1.1 | 11B | None | +| ERNIE-Image | ERNIE text encoder | not specified | Flux.2 VAE | +| Flux.1 | CLIP-L/14 + T5 XXL v1.1 | 123M + 11B | AutoencoderKL | +| Flux.2 | Mistral-Small-3.1-24B | 24B | Flux.2 VAE | +| HeartMuLa | None | N/A | HeartCodec tokens | +| HiDream | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 + Llama | 123M + 694M + 11B + not specified | AutoencoderKL | +| Hunyuan Video | Hunyuan LLM | not specified | Hunyuan Video 3D VAE | +| Ideogram 4 | Qwen3-VL-8B-Instruct | 8B | Ideogram AutoEncoder | +| Kandinsky 5.0 Image | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Flux VAE (AutoencoderKL) | +| Kandinsky 5.0 Video | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Hunyuan Video VAE | +| Kwai Kolors | ChatGLM-6B | 6B | AutoencoderKL | +| Krea2 | Qwen3VL | not specified | Qwen Image VAE | +| LongCat Image | Qwen2.5-VL | 7B | AutoencoderKL | +| LongCat Video | Qwen2.5-VL | 7B | Wan VAE | +| LTX Video | T5 XXL v1.1 | 11B | LTX Video VAE | +| LTX Video 2 | Gemma3 | not specified | LTX Video 2 VAE | +| Lumina2 | Gemma2 | 2B | AutoencoderKL | +| Mage-Flow | Qwen3-VL | not specified | Mage-VAE | +| OmniGen | Integrated OmniGen encoder | not specified | AutoencoderKL | +| PixArt Sigma | T5 XXL v1.1 | 11B | AutoencoderKL | +| Qwen Image | Qwen2.5-VL | 7B | Qwen Image VAE | +| Sana | Gemma2 2B-IT | 2B | Sana AutoencoderDC | +| Sana Video | Gemma 2 | 2B | Wan VAE | +| SD 1.x/2.x (Legacy) | CLIP-L/14 | 123M | AutoencoderKL | +| Stable Diffusion 3 | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 | 123M + 694M + 11B | AutoencoderKL | +| Stable Diffusion XL | CLIP-L/14 + CLIP-G/14 | 123M + 694M | AutoencoderKL | +| Stable Cascade (Stage C) | CLIP-ViT-bigG-14 | 694M | Stable Cascade Stage C VAE | +| Wan Video | UMT5 | not specified | Wan VAE | +| Wan S2V | UMT5 | not specified | Wan VAE | +| Z-Image | Qwen3 4B | 4B | AutoencoderKL | +| Z-Image Omni | Qwen3 4B | 4B | AutoencoderKL | +| ZLab I1 | T5Gemma 2B | 2B | AutoencoderKL | + +
+ +*✓ = サポート、✓* = サポートされていますが、full-rank training では通常 DeepSpeed/FSDP2 が必要、✗ = 未サポート。Ref Inputs は既存の reference/edit/I2V conditioning path を示します。`opt` は任意、`req` は edit/I2V flavour で必須です。* +*TwinFlow は `twinflow_enabled=true` のとき native support です。diffusion models では `diff2flow_enabled=true` と `twinflow_allow_diff2flow=true` も必要です。Self-Flow は CREPA self-flow support です。LayerSync は alignment 用 hidden states を公開する backbone を示します。* ### 高速パス: Z-Image TurboとFlux Schnell diff --git a/documentation/QUICKSTART.md b/documentation/QUICKSTART.md index 63cdbe2a7..240dacf64 100644 --- a/documentation/QUICKSTART.md +++ b/documentation/QUICKSTART.md @@ -2,55 +2,291 @@ **Note**: For more advanced configurations, see the [tutorial](TUTORIAL.md) and [options reference](OPTIONS.md). +## Model Quickstart Guides + +| Model | Parameters | Guide | +| --- | --- | --- | +| ACE-Step | 3.5B | [ACE_STEP.md](/documentation/quickstart/ACE_STEP.md) | +| Anima | Not specified | No dedicated guide | +| Auraflow | 6B | [AURAFLOW.md](/documentation/quickstart/AURAFLOW.md) | +| Boogu-Image | Not specified | [BOOGU_IMAGE.md](/documentation/quickstart/BOOGU_IMAGE.md) | +| Chroma 1 | 8.9B | [CHROMA.md](/documentation/quickstart/CHROMA.md) | +| Cosmos2 | 2B-14B | [COSMOS2IMAGE.md](/documentation/quickstart/COSMOS2IMAGE.md) | +| Cosmos3 | 16B-65B | [COSMOS3.md](/documentation/quickstart/COSMOS3.md) | +| DeepFloyd IF | 0.4B-4.3B stages | No dedicated guide | +| ERNIE-Image | Not specified | [ERNIE.md](/documentation/quickstart/ERNIE.md) | +| Flux.1 | 8B-12B | [FLUX.md](/documentation/quickstart/FLUX.md)
[FLUX_KONTEXT.md](/documentation/quickstart/FLUX_KONTEXT.md) | +| Flux.2 | 4B-32B | [FLUX2.md](/documentation/quickstart/FLUX2.md) | +| HeartMuLa | 3B | [HEARTMULA.md](/documentation/quickstart/HEARTMULA.md) | +| HiDream | 17B (8.5B MoE) | [HIDREAM.md](/documentation/quickstart/HIDREAM.md) | +| Hunyuan Video | 8.3B | [HUNYUANVIDEO.md](/documentation/quickstart/HUNYUANVIDEO.md) | +| Ideogram 4 | 9B | [IDEOGRAM4.md](/documentation/quickstart/IDEOGRAM4.md) | +| Kandinsky 5.0 Image | 6B (lite) | [KANDINSKY5_IMAGE.md](/documentation/quickstart/KANDINSKY5_IMAGE.md) | +| Kandinsky 5.0 Video | 2B lite, 19B pro | [KANDINSKY5_VIDEO.md](/documentation/quickstart/KANDINSKY5_VIDEO.md) | +| Kwai Kolors | 2.7B | [KOLORS.md](/documentation/quickstart/KOLORS.md) | +| Krea2 | Not specified | [KREA2.md](/documentation/quickstart/KREA2.md) | +| LongCat Image | 6B | [LONGCAT_IMAGE.md](/documentation/quickstart/LONGCAT_IMAGE.md)
[LONGCAT_EDIT.md](/documentation/quickstart/LONGCAT_EDIT.md) | +| LongCat Video | 13.6B | [LONGCAT_VIDEO.md](/documentation/quickstart/LONGCAT_VIDEO.md)
[LONGCAT_VIDEO_EDIT.md](/documentation/quickstart/LONGCAT_VIDEO_EDIT.md) | +| LTX Video | ~2.5B | [LTXVIDEO.md](/documentation/quickstart/LTXVIDEO.md) | +| LTX Video 2 | 19B | [LTXVIDEO2.md](/documentation/quickstart/LTXVIDEO2.md) | +| Lumina2 | 2B | [LUMINA2.md](/documentation/quickstart/LUMINA2.md) | +| Mage-Flow | 4B | [MAGEFLOW.md](/documentation/quickstart/MAGEFLOW.md) | +| OmniGen | 3.8B | [OMNIGEN.md](/documentation/quickstart/OMNIGEN.md) | +| PixArt Sigma | 0.6B-0.9B | [SIGMA.md](/documentation/quickstart/SIGMA.md) | +| Qwen Image | 20B | [QWEN_IMAGE.md](/documentation/quickstart/QWEN_IMAGE.md)
[QWEN_EDIT.md](/documentation/quickstart/QWEN_EDIT.md) | +| Sana | 0.6B-4.8B | [SANA.md](/documentation/quickstart/SANA.md) | +| Sana Video | 2B | [SANAVIDEO.md](/documentation/quickstart/SANAVIDEO.md) | +| SD 1.x/2.x (Legacy) | 0.9B | No dedicated guide | +| Stable Diffusion 3 | 2B-8B | [SD3.md](/documentation/quickstart/SD3.md) | +| Stable Diffusion XL | 3.5B | [SDXL.md](/documentation/quickstart/SDXL.md) | +| Stable Cascade (Stage C) | 1B, 3.6B prior | [STABLE_CASCADE_C.md](/documentation/quickstart/STABLE_CASCADE_C.md) | +| Wan Video | 1.3B-14B | [WAN.md](/documentation/quickstart/WAN.md) | +| Wan S2V | 14B | [WAN_S2V.md](/documentation/quickstart/WAN_S2V.md) | +| Z-Image | 6B | [ZIMAGE.md](/documentation/quickstart/ZIMAGE.md) | +| Z-Image Omni | 6B | [ZIMAGE.md](/documentation/quickstart/ZIMAGE.md) | +| ZLab I1 | 3B | [ZLAB_i1.md](/documentation/quickstart/ZLAB_i1.md) | + ## Feature Compatibility -For the complete and most accurate feature matrix, refer to the [main README](https://github.com/bghira/SimpleTuner#model-architecture-support). +The complete compatibility matrix is split by feature area so each table stays readable. -## Model Quickstart Guides +
+Training support + +| Model | PEFT LoRA | LyCORIS | Full-Rank | ControlNet | Ref Inputs | +| --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓* | ✗ | ✗ | +| Anima | ✓ | ✓ | ✓* | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓* | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| Chroma 1 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓* | ✗ | audio opt | +| DeepFloyd IF | ✓ | ✓ | ✓ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Flux.1 | ✓ | ✓ | ✓* | ✓ | ✓ opt (Kontext) | +| Flux.2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| HeartMuLa | ✓ | ✓ | ✓* | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓* | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Ideogram 4 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓* | ✗ | ✓ I2I | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Kwai Kolors | ✓ | ✓ | ✓ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| LongCat Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| LongCat Video | ✓ | ✓ | ✓* | ✗ | ✓ opt/edit | +| LTX Video | ✓ | ✓ | ✓ | ✗ | ✓ I2V | +| LTX Video 2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| OmniGen | ✓ | ✓ | ✓ | ✗ | ✗ | +| PixArt Sigma | ✗ | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| Sana | ✗ | ✓ | ✓ | ✗ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓* | ✓ | ✗ | +| Stable Diffusion XL | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓* | ✗ | ✗ | +| Wan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V/VACE | +| Wan S2V | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image Omni | ✓ | ✓ | ✓* | ✗ | ✓ opt (Edit) | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | ✗ | + +
+ +
+Precision level support + +| Model | Quantization | Mixed Precision | +| --- | --- | --- | +| ACE-Step | int8 optional | bf16 | +| Anima | not specified | bf16 | +| Auraflow | int8/fp8/nf4 optional | bf16 | +| Boogu-Image | fp8 optional | bf16 | +| Chroma 1 | int8/fp8/nf4 optional | bf16 | +| Cosmos2 | int8 optional | bf16 | +| Cosmos3 | no_change first; int8 optional | bf16 | +| DeepFloyd IF | not recommended | bf16 | +| ERNIE-Image | int8 optional | bf16 | +| Flux.1 | int8/fp8/nf4 optional | bf16 | +| Flux.2 | int8/fp8/nf4 optional | bf16 | +| HeartMuLa | int8 optional | bf16 | +| HiDream | int8/fp8/nf4 optional | bf16 | +| Hunyuan Video | int8 optional | bf16 | +| Ideogram 4 | fp8 default, nf4 optional | bf16 | +| Kandinsky 5.0 Image | int8 optional | bf16 | +| Kandinsky 5.0 Video | int8 optional | bf16 | +| Kwai Kolors | not recommended | bf16 | +| Krea2 | int8 optional | bf16 | +| LongCat Image | int8/fp8 optional | bf16 | +| LongCat Video | int8/fp8 optional | bf16 | +| LTX Video | int8/fp8 optional | bf16 | +| LTX Video 2 | int8/fp8 optional | bf16 | +| Lumina2 | int8 optional | bf16 | +| Mage-Flow | fp8 optional | bf16 | +| OmniGen | int8/fp8 optional | bf16 | +| PixArt Sigma | int8 optional | bf16 | +| Qwen Image | required (int8/nf4) | bf16 | +| Sana | int8 optional | bf16 | +| Sana Video | not recommended for full | bf16 | +| SD 1.x/2.x (Legacy) | int8/nf4 optional | bf16 | +| Stable Diffusion 3 | int8/fp8/nf4 optional | bf16 | +| Stable Diffusion XL | int8/nf4 optional | bf16 | +| Stable Cascade (Stage C) | not supported | fp32 required | +| Wan Video | int8 optional | bf16 | +| Wan S2V | int8 optional | bf16 | +| Z-Image | int8 optional | bf16 | +| Z-Image Omni | int8 optional | bf16 | +| ZLab I1 | int8 optional | bf16 | + +
+ +
+Checkpointing granularity + +| Model | Gradient Checkpoint | Interval | Segment Stride | Attention Offload | +| --- | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓ | ✗ | +| Anima | ✓ | ✗ | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓ | ✗ | +| Chroma 1 | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓ | ✗ | +| DeepFloyd IF | ✓ | ✗ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓ | ✗ | +| Flux.1 | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | ✓ | ✗ | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | ✓ | ✓ | ✓ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | ✓ | ✗ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | ✓ | ✓ | ✓ | ✓ | +| LTX Video | ✓ | ✓ | ✓ | ✗ | +| LTX Video 2 | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓ | ✓ | +| OmniGen | ✓ | ✗ | ✗ | ✗ | +| PixArt Sigma | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓ | ✗ | +| Sana | ✓ | ✓ | ✓ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✗ | ✗ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | ✓ | ✗ | ✗ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓ | ✗ | +| Wan Video | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | ✓ | ✓ | ✓ | ✗ | +| Z-Image | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | ✓ | ✗ | ✗ | ✗ | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | + +
+ +
+Flow, distillation, and alignment + +| Model | Prediction | Flow Shift | TwinFlow | Self-Flow | LayerSync | Sliders | +| --- | --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Anima | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Auraflow | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Boogu-Image | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Chroma 1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | sample | ✗ | ✗ | ✓ | ✓ | ✓ | +| Cosmos3 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| DeepFloyd IF | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| ERNIE-Image | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Flux.1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | autoregressive next-token | ✗ | ✗ | ✗ | ✗ | ✗ | +| HiDream | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Hunyuan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Kandinsky 5.0 Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Krea2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video 2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Mage-Flow | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| OmniGen | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| PixArt Sigma | epsilon | ✗ | ✗ | ✓ | ✓ | ✓ | +| Qwen Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| SD 1.x/2.x (Legacy) | epsilon / v-pred | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Diffusion 3 | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Cascade (Stage C) | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Wan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Z-Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| ZLab I1 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | + +
+ +
+Text encoders and VAE types + +| Model | Text Encoders | Text Encoder Params | VAE | +| --- | --- | --- | --- | +| ACE-Step | UMT5 Encoder | 0.6B | Music DCAE | +| Anima | Qwen3 0.6B | 0.6B | Qwen Image VAE | +| Auraflow | Pile T5 | not specified | AutoencoderKL | +| Boogu-Image | Qwen3-VL | not specified | AutoencoderKL | +| Chroma 1 | T5 XXL v1.1 | 11B | AutoencoderKL | +| Cosmos2 | T5 11B | 11B | Wan VAE | +| Cosmos3 | Cosmos3 reasoner | not specified | Wan/Cosmos VAE | +| DeepFloyd IF | T5 XXL v1.1 | 11B | None | +| ERNIE-Image | ERNIE text encoder | not specified | Flux.2 VAE | +| Flux.1 | CLIP-L/14 + T5 XXL v1.1 | 123M + 11B | AutoencoderKL | +| Flux.2 | Mistral-Small-3.1-24B | 24B | Flux.2 VAE | +| HeartMuLa | None | N/A | HeartCodec tokens | +| HiDream | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 + Llama | 123M + 694M + 11B + not specified | AutoencoderKL | +| Hunyuan Video | Hunyuan LLM | not specified | Hunyuan Video 3D VAE | +| Ideogram 4 | Qwen3-VL-8B-Instruct | 8B | Ideogram AutoEncoder | +| Kandinsky 5.0 Image | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Flux VAE (AutoencoderKL) | +| Kandinsky 5.0 Video | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Hunyuan Video VAE | +| Kwai Kolors | ChatGLM-6B | 6B | AutoencoderKL | +| Krea2 | Qwen3VL | not specified | Qwen Image VAE | +| LongCat Image | Qwen2.5-VL | 7B | AutoencoderKL | +| LongCat Video | Qwen2.5-VL | 7B | Wan VAE | +| LTX Video | T5 XXL v1.1 | 11B | LTX Video VAE | +| LTX Video 2 | Gemma3 | not specified | LTX Video 2 VAE | +| Lumina2 | Gemma2 | 2B | AutoencoderKL | +| Mage-Flow | Qwen3-VL | not specified | Mage-VAE | +| OmniGen | Integrated OmniGen encoder | not specified | AutoencoderKL | +| PixArt Sigma | T5 XXL v1.1 | 11B | AutoencoderKL | +| Qwen Image | Qwen2.5-VL | 7B | Qwen Image VAE | +| Sana | Gemma2 2B-IT | 2B | Sana AutoencoderDC | +| Sana Video | Gemma 2 | 2B | Wan VAE | +| SD 1.x/2.x (Legacy) | CLIP-L/14 | 123M | AutoencoderKL | +| Stable Diffusion 3 | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 | 123M + 694M + 11B | AutoencoderKL | +| Stable Diffusion XL | CLIP-L/14 + CLIP-G/14 | 123M + 694M | AutoencoderKL | +| Stable Cascade (Stage C) | CLIP-ViT-bigG-14 | 694M | Stable Cascade Stage C VAE | +| Wan Video | UMT5 | not specified | Wan VAE | +| Wan S2V | UMT5 | not specified | Wan VAE | +| Z-Image | Qwen3 4B | 4B | AutoencoderKL | +| Z-Image Omni | Qwen3 4B | 4B | AutoencoderKL | +| ZLab I1 | T5Gemma 2B | 2B | AutoencoderKL | + +
-| Model | Params | PEFT LoRA | Full-Rank | Quantization | Mixed Precision | Grad Checkpoint | Flow Shift | TwinFlow | Self-Flow | LayerSync | Ref Inputs | ControlNet | Sliders† | Guide | -| --- | --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | --- | -| PixArt Sigma | 0.6B–0.9B | ✗ | ✓ | int8 optional | bf16 | ✓ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ | ✓ | [SIGMA.md](/documentation/quickstart/SIGMA.md) | -| NVLabs Sana | 1.6B–4.8B | ✗ | ✓ | int8 optional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [SANA.md](/documentation/quickstart/SANA.md) | -| Kwai Kolors | 2.7B | ✓ | ✓ | not recommended | bf16 | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [KOLORS.md](/documentation/quickstart/KOLORS.md) | -| Stable Diffusion 3 | 2B–8B | ✓ | ✓ | int8/fp8/nf4 optional | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [SD3.md](/documentation/quickstart/SD3.md) | -| Flux.1 | 8B–12B | ✓ | ✓* | int8/fp8/nf4 optional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [FLUX.md](/documentation/quickstart/FLUX.md) | -| Flux.2 | 32B | ✓ | ✓* | int8/fp8/nf4 optional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [FLUX2.md](/documentation/quickstart/FLUX2.md) | -| Flux Kontext | 8B–12B | ✓ | ✓* | int8/fp8/nf4 optional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✓ | ✓ | [FLUX_KONTEXT.md](/documentation/quickstart/FLUX_KONTEXT.md) | -| Z-Image Turbo | 6B | ✓ | ✓* | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZIMAGE.md](/documentation/quickstart/ZIMAGE.md) | -| Krea2 | - | ✓ | ✓* | int8 optional | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✓ opt | ✗ | ✓ | [KREA2.md](/documentation/quickstart/KREA2.md) | -| Boogu-Image 0.1 | - | ✓ | ✓* | fp8 optional | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ edit | ✗ | ✓ | [BOOGU_IMAGE.md](/documentation/quickstart/BOOGU_IMAGE.md) | -| zlab i1 | 3B | ✓ | ✓ | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZLAB_i1.md](/documentation/quickstart/ZLAB_i1.md) | -| Ideogram 4 | 9B | ✓ | ✓* | fp8 default, nf4 optional | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [IDEOGRAM4.md](/documentation/quickstart/IDEOGRAM4.md) | -| ACE-Step | 3.5B | ✓ | ✓* | int8 optional | bf16 | ✓ | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✓ | [ACE_STEP.md](/documentation/quickstart/ACE_STEP.md) | -| Chroma 1 | 8.9B | ✓ | ✓* | int8/fp8/nf4 optional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [CHROMA.md](/documentation/quickstart/CHROMA.md) | -| Auraflow | 6B | ✓ | ✓* | int8/fp8/nf4 optional | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [AURAFLOW.md](/documentation/quickstart/AURAFLOW.md) | -| HiDream I1 | 17B (8.5B MoE) | ✓ | ✓* | int8/fp8/nf4 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [HIDREAM.md](/documentation/quickstart/HIDREAM.md) | -| OmniGen | 3.8B | ✓ | ✓ | int8/fp8 optional | bf16 | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | [OMNIGEN.md](/documentation/quickstart/OMNIGEN.md) | -| Stable Diffusion XL | 2.6B | ✓ | ✓ | not recommended | bf16 | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | [SDXL.md](/documentation/quickstart/SDXL.md) | -| Lumina2 | 2B | ✓ | ✓ | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | [LUMINA2.md](/documentation/quickstart/LUMINA2.md) | -| Cosmos2 | 2B | ✓ | ✓ | not recommended | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [COSMOS2IMAGE.md](/documentation/quickstart/COSMOS2IMAGE.md) | -| Cosmos3 | 16B-65B | ✓ | ✓* | no_change first | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | audio opt | ✗ | ✓ | [COSMOS3.md](/documentation/quickstart/COSMOS3.md) | -| LTX Video | ~2.5B | ✓ | ✓ | int8/fp8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [LTXVIDEO.md](/documentation/quickstart/LTXVIDEO.md) | -| LTX Video 2 | 19B | ✓ | ✓* | int8/fp8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LTXVIDEO2.md](/documentation/quickstart/LTXVIDEO2.md) | -| Hunyuan Video 1.5 | 8.3B | ✓ | ✓* | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [HUNYUANVIDEO.md](/documentation/quickstart/HUNYUANVIDEO.md) | -| Wan 2.x | 1.3B–14B | ✓ | ✓* | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN.md](/documentation/quickstart/WAN.md) | -| Wan 2.2 S2V | 14B | ✓ | ✓* | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN_S2V.md](/documentation/quickstart/WAN_S2V.md) | -| Qwen Image | 20B | ✓ | ✓* | **required** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [QWEN_IMAGE.md](/documentation/quickstart/QWEN_IMAGE.md) | -| Qwen Image Edit | 20B | ✓ | ✓* | **required** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [QWEN_EDIT.md](/documentation/quickstart/QWEN_EDIT.md) | -| Stable Cascade (C) | 1B, 3.6B prior | ✓ | ✓* | not supported | fp32 (required) | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [STABLE_CASCADE_C.md](/documentation/quickstart/STABLE_CASCADE_C.md) | -| Kandinsky 5.0 Image | 6B (lite) | ✓ | ✓* | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ I2I | ✗ | ✓ | [KANDINSKY5_IMAGE.md](/documentation/quickstart/KANDINSKY5_IMAGE.md) | -| Kandinsky 5.0 Video | 2B (lite), 19B (pro) | ✓ | ✓* | int8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [KANDINSKY5_VIDEO.md](/documentation/quickstart/KANDINSKY5_VIDEO.md) | -| LongCat-Video | 13.6B | ✓ | ✓* | int8/fp8 optional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LONGCAT_VIDEO.md](/documentation/quickstart/LONGCAT_VIDEO.md) | -| LongCat-Video Edit | 13.6B | ✓ | ✓* | int8/fp8 optional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_VIDEO_EDIT.md](/documentation/quickstart/LONGCAT_VIDEO_EDIT.md) | -| LongCat-Image | 6B | ✓ | ✓* | int8/fp8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [LONGCAT_IMAGE.md](/documentation/quickstart/LONGCAT_IMAGE.md) | -| LongCat-Image Edit | 6B | ✓ | ✓* | int8/fp8 optional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_EDIT.md](/documentation/quickstart/LONGCAT_EDIT.md) | - -*✓ = supported, ✓* = requires DeepSpeed/FSDP2 for full-rank, ✗ = not supported, `✓+` indicates checkpointing is recommended due to VRAM pressure. Ref Inputs marks existing reference/edit/I2V conditioning paths; `opt` means optional, `req` means the edit/I2V flavour requires it. TwinFlow ✓ means native support when `twinflow_enabled=true` (diffusion models need `diff2flow_enabled+twinflow_allow_diff2flow`). Self-Flow ✓ means native support for `crepa_enabled=true` with `crepa_feature_source=self_flow`, `use_ema=true`, and `crepa_teacher_block_index` set. LayerSync ✓ means the backbone exposes transformer hidden states for self-alignment; ✗ marks UNet-style backbones without that buffer. †Sliders apply to LoRA and LyCORIS (including full-rank LyCORIS "full"). All models support LyCORIS.* - -> ℹ️ Wan quickstart includes 2.1 + 2.2 stage presets and the time-embedding toggle. Flux Kontext covers editing workflows built atop Flux.1. - -> ⚠️ These quickstarts are living documents. Expect occasional updates as new models land or training recipes improve. +*✓ = supported, ✓* = supported but usually requires DeepSpeed/FSDP2 for full-rank training, ✗ = not supported. Ref Inputs marks existing reference/edit/I2V conditioning paths; `opt` means optional and `req` means the edit/I2V flavour requires it.* +*TwinFlow support is native when `twinflow_enabled=true`; diffusion models still require `diff2flow_enabled=true` plus `twinflow_allow_diff2flow=true`. Self-Flow refers to CREPA self-flow support. LayerSync marks backbones that expose hidden states for alignment.* ### Fast paths: Z-Image Turbo & Flux Schnell diff --git a/documentation/QUICKSTART.pt-BR.md b/documentation/QUICKSTART.pt-BR.md index 753962ece..3daf9f2ef 100644 --- a/documentation/QUICKSTART.pt-BR.md +++ b/documentation/QUICKSTART.pt-BR.md @@ -2,55 +2,291 @@ **Nota**: Para configurações mais avançadas, veja o [tutorial](TUTORIAL.md) e a [referência de opções](OPTIONS.md). +## Guias de início rápido por modelo + +| Modelo | Parâmetros | Guia | +| --- | --- | --- | +| ACE-Step | 3.5B | [ACE_STEP.pt-BR.md](quickstart/ACE_STEP.pt-BR.md) | +| Anima | Not specified | Sem guia dedicada | +| Auraflow | 6B | [AURAFLOW.pt-BR.md](quickstart/AURAFLOW.pt-BR.md) | +| Boogu-Image | Not specified | [BOOGU_IMAGE.pt-BR.md](quickstart/BOOGU_IMAGE.pt-BR.md) | +| Chroma 1 | 8.9B | [CHROMA.pt-BR.md](quickstart/CHROMA.pt-BR.md) | +| Cosmos2 | 2B-14B | [COSMOS2IMAGE.pt-BR.md](quickstart/COSMOS2IMAGE.pt-BR.md) | +| Cosmos3 | 16B-65B | [COSMOS3.pt-BR.md](quickstart/COSMOS3.pt-BR.md) | +| DeepFloyd IF | 0.4B-4.3B stages | Sem guia dedicada | +| ERNIE-Image | Not specified | [ERNIE.pt-BR.md](quickstart/ERNIE.pt-BR.md) | +| Flux.1 | 8B-12B | [FLUX.pt-BR.md](quickstart/FLUX.pt-BR.md)
[FLUX_KONTEXT.pt-BR.md](quickstart/FLUX_KONTEXT.pt-BR.md) | +| Flux.2 | 4B-32B | [FLUX2.pt-BR.md](quickstart/FLUX2.pt-BR.md) | +| HeartMuLa | 3B | [HEARTMULA.pt-BR.md](quickstart/HEARTMULA.pt-BR.md) | +| HiDream | 17B (8.5B MoE) | [HIDREAM.pt-BR.md](quickstart/HIDREAM.pt-BR.md) | +| Hunyuan Video | 8.3B | [HUNYUANVIDEO.pt-BR.md](quickstart/HUNYUANVIDEO.pt-BR.md) | +| Ideogram 4 | 9B | [IDEOGRAM4.pt-BR.md](quickstart/IDEOGRAM4.pt-BR.md) | +| Kandinsky 5.0 Image | 6B (lite) | [KANDINSKY5_IMAGE.pt-BR.md](quickstart/KANDINSKY5_IMAGE.pt-BR.md) | +| Kandinsky 5.0 Video | 2B lite, 19B pro | [KANDINSKY5_VIDEO.pt-BR.md](quickstart/KANDINSKY5_VIDEO.pt-BR.md) | +| Kwai Kolors | 2.7B | [KOLORS.pt-BR.md](quickstart/KOLORS.pt-BR.md) | +| Krea2 | Not specified | [KREA2.pt-BR.md](quickstart/KREA2.pt-BR.md) | +| LongCat Image | 6B | [LONGCAT_IMAGE.pt-BR.md](quickstart/LONGCAT_IMAGE.pt-BR.md)
[LONGCAT_EDIT.pt-BR.md](quickstart/LONGCAT_EDIT.pt-BR.md) | +| LongCat Video | 13.6B | [LONGCAT_VIDEO.pt-BR.md](quickstart/LONGCAT_VIDEO.pt-BR.md)
[LONGCAT_VIDEO_EDIT.pt-BR.md](quickstart/LONGCAT_VIDEO_EDIT.pt-BR.md) | +| LTX Video | ~2.5B | [LTXVIDEO.pt-BR.md](quickstart/LTXVIDEO.pt-BR.md) | +| LTX Video 2 | 19B | [LTXVIDEO2.pt-BR.md](quickstart/LTXVIDEO2.pt-BR.md) | +| Lumina2 | 2B | [LUMINA2.pt-BR.md](quickstart/LUMINA2.pt-BR.md) | +| Mage-Flow | 4B | [MAGEFLOW.pt-BR.md](quickstart/MAGEFLOW.pt-BR.md) | +| OmniGen | 3.8B | [OMNIGEN.pt-BR.md](quickstart/OMNIGEN.pt-BR.md) | +| PixArt Sigma | 0.6B-0.9B | [SIGMA.pt-BR.md](quickstart/SIGMA.pt-BR.md) | +| Qwen Image | 20B | [QWEN_IMAGE.pt-BR.md](quickstart/QWEN_IMAGE.pt-BR.md)
[QWEN_EDIT.pt-BR.md](quickstart/QWEN_EDIT.pt-BR.md) | +| Sana | 0.6B-4.8B | [SANA.pt-BR.md](quickstart/SANA.pt-BR.md) | +| Sana Video | 2B | [SANAVIDEO.pt-BR.md](quickstart/SANAVIDEO.pt-BR.md) | +| SD 1.x/2.x (Legacy) | 0.9B | Sem guia dedicada | +| Stable Diffusion 3 | 2B-8B | [SD3.pt-BR.md](quickstart/SD3.pt-BR.md) | +| Stable Diffusion XL | 3.5B | [SDXL.pt-BR.md](quickstart/SDXL.pt-BR.md) | +| Stable Cascade (Stage C) | 1B, 3.6B prior | [STABLE_CASCADE_C.pt-BR.md](quickstart/STABLE_CASCADE_C.pt-BR.md) | +| Wan Video | 1.3B-14B | [WAN.pt-BR.md](quickstart/WAN.pt-BR.md) | +| Wan S2V | 14B | [WAN_S2V.pt-BR.md](quickstart/WAN_S2V.pt-BR.md) | +| Z-Image | 6B | [ZIMAGE.pt-BR.md](quickstart/ZIMAGE.pt-BR.md) | +| Z-Image Omni | 6B | [ZIMAGE.pt-BR.md](quickstart/ZIMAGE.pt-BR.md) | +| ZLab I1 | 3B | [ZLAB_i1.pt-BR.md](quickstart/ZLAB_i1.pt-BR.md) | + ## Compatibilidade de recursos -Para a matriz completa e mais precisa de recursos, consulte o [README principal](https://github.com/bghira/SimpleTuner#model-architecture-support). +A matriz completa de compatibilidade é dividida por área de recurso para manter cada tabela legível. -## Guias de início rápido por modelo +
+Suporte de treinamento + +| Modelo | PEFT LoRA | LyCORIS | Full-Rank | ControlNet | Ref Inputs | +| --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓* | ✗ | ✗ | +| Anima | ✓ | ✓ | ✓* | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓* | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| Chroma 1 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓* | ✗ | audio opt | +| DeepFloyd IF | ✓ | ✓ | ✓ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Flux.1 | ✓ | ✓ | ✓* | ✓ | ✓ opt (Kontext) | +| Flux.2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| HeartMuLa | ✓ | ✓ | ✓* | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓* | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Ideogram 4 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓* | ✗ | ✓ I2I | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Kwai Kolors | ✓ | ✓ | ✓ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| LongCat Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| LongCat Video | ✓ | ✓ | ✓* | ✗ | ✓ opt/edit | +| LTX Video | ✓ | ✓ | ✓ | ✗ | ✓ I2V | +| LTX Video 2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| OmniGen | ✓ | ✓ | ✓ | ✗ | ✗ | +| PixArt Sigma | ✗ | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| Sana | ✗ | ✓ | ✓ | ✗ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓* | ✓ | ✗ | +| Stable Diffusion XL | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓* | ✗ | ✗ | +| Wan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V/VACE | +| Wan S2V | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image Omni | ✓ | ✓ | ✓* | ✗ | ✓ opt (Edit) | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | ✗ | + +
+ +
+Suporte a níveis de precisão + +| Modelo | Quantização | Precisão mista | +| --- | --- | --- | +| ACE-Step | int8 optional | bf16 | +| Anima | not specified | bf16 | +| Auraflow | int8/fp8/nf4 optional | bf16 | +| Boogu-Image | fp8 optional | bf16 | +| Chroma 1 | int8/fp8/nf4 optional | bf16 | +| Cosmos2 | int8 optional | bf16 | +| Cosmos3 | no_change first; int8 optional | bf16 | +| DeepFloyd IF | not recommended | bf16 | +| ERNIE-Image | int8 optional | bf16 | +| Flux.1 | int8/fp8/nf4 optional | bf16 | +| Flux.2 | int8/fp8/nf4 optional | bf16 | +| HeartMuLa | int8 optional | bf16 | +| HiDream | int8/fp8/nf4 optional | bf16 | +| Hunyuan Video | int8 optional | bf16 | +| Ideogram 4 | fp8 default, nf4 optional | bf16 | +| Kandinsky 5.0 Image | int8 optional | bf16 | +| Kandinsky 5.0 Video | int8 optional | bf16 | +| Kwai Kolors | not recommended | bf16 | +| Krea2 | int8 optional | bf16 | +| LongCat Image | int8/fp8 optional | bf16 | +| LongCat Video | int8/fp8 optional | bf16 | +| LTX Video | int8/fp8 optional | bf16 | +| LTX Video 2 | int8/fp8 optional | bf16 | +| Lumina2 | int8 optional | bf16 | +| Mage-Flow | fp8 optional | bf16 | +| OmniGen | int8/fp8 optional | bf16 | +| PixArt Sigma | int8 optional | bf16 | +| Qwen Image | required (int8/nf4) | bf16 | +| Sana | int8 optional | bf16 | +| Sana Video | not recommended for full | bf16 | +| SD 1.x/2.x (Legacy) | int8/nf4 optional | bf16 | +| Stable Diffusion 3 | int8/fp8/nf4 optional | bf16 | +| Stable Diffusion XL | int8/nf4 optional | bf16 | +| Stable Cascade (Stage C) | not supported | fp32 required | +| Wan Video | int8 optional | bf16 | +| Wan S2V | int8 optional | bf16 | +| Z-Image | int8 optional | bf16 | +| Z-Image Omni | int8 optional | bf16 | +| ZLab I1 | int8 optional | bf16 | + +
+ +
+Granularidade de checkpointing + +| Modelo | Gradient Checkpoint | Interval | Segment Stride | Attention Offload | +| --- | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓ | ✗ | +| Anima | ✓ | ✗ | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓ | ✗ | +| Chroma 1 | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓ | ✗ | +| DeepFloyd IF | ✓ | ✗ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓ | ✗ | +| Flux.1 | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | ✓ | ✗ | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | ✓ | ✓ | ✓ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | ✓ | ✗ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | ✓ | ✓ | ✓ | ✓ | +| LTX Video | ✓ | ✓ | ✓ | ✗ | +| LTX Video 2 | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓ | ✓ | +| OmniGen | ✓ | ✗ | ✗ | ✗ | +| PixArt Sigma | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓ | ✗ | +| Sana | ✓ | ✓ | ✓ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✗ | ✗ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | ✓ | ✗ | ✗ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓ | ✗ | +| Wan Video | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | ✓ | ✓ | ✓ | ✗ | +| Z-Image | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | ✓ | ✗ | ✗ | ✗ | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | + +
+ +
+Flow, destilação e alinhamento + +| Modelo | Prediction | Flow Shift | TwinFlow | Self-Flow | LayerSync | Sliders | +| --- | --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Anima | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Auraflow | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Boogu-Image | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Chroma 1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | sample | ✗ | ✗ | ✓ | ✓ | ✓ | +| Cosmos3 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| DeepFloyd IF | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| ERNIE-Image | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Flux.1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | autoregressive next-token | ✗ | ✗ | ✗ | ✗ | ✗ | +| HiDream | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Hunyuan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Kandinsky 5.0 Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Krea2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video 2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Mage-Flow | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| OmniGen | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| PixArt Sigma | epsilon | ✗ | ✗ | ✓ | ✓ | ✓ | +| Qwen Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| SD 1.x/2.x (Legacy) | epsilon / v-pred | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Diffusion 3 | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Cascade (Stage C) | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Wan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Z-Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| ZLab I1 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | + +
+ +
+Text encoders e tipos de VAE + +| Modelo | Text Encoders | Text Encoder Params | VAE | +| --- | --- | --- | --- | +| ACE-Step | UMT5 Encoder | 0.6B | Music DCAE | +| Anima | Qwen3 0.6B | 0.6B | Qwen Image VAE | +| Auraflow | Pile T5 | not specified | AutoencoderKL | +| Boogu-Image | Qwen3-VL | not specified | AutoencoderKL | +| Chroma 1 | T5 XXL v1.1 | 11B | AutoencoderKL | +| Cosmos2 | T5 11B | 11B | Wan VAE | +| Cosmos3 | Cosmos3 reasoner | not specified | Wan/Cosmos VAE | +| DeepFloyd IF | T5 XXL v1.1 | 11B | None | +| ERNIE-Image | ERNIE text encoder | not specified | Flux.2 VAE | +| Flux.1 | CLIP-L/14 + T5 XXL v1.1 | 123M + 11B | AutoencoderKL | +| Flux.2 | Mistral-Small-3.1-24B | 24B | Flux.2 VAE | +| HeartMuLa | None | N/A | HeartCodec tokens | +| HiDream | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 + Llama | 123M + 694M + 11B + not specified | AutoencoderKL | +| Hunyuan Video | Hunyuan LLM | not specified | Hunyuan Video 3D VAE | +| Ideogram 4 | Qwen3-VL-8B-Instruct | 8B | Ideogram AutoEncoder | +| Kandinsky 5.0 Image | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Flux VAE (AutoencoderKL) | +| Kandinsky 5.0 Video | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Hunyuan Video VAE | +| Kwai Kolors | ChatGLM-6B | 6B | AutoencoderKL | +| Krea2 | Qwen3VL | not specified | Qwen Image VAE | +| LongCat Image | Qwen2.5-VL | 7B | AutoencoderKL | +| LongCat Video | Qwen2.5-VL | 7B | Wan VAE | +| LTX Video | T5 XXL v1.1 | 11B | LTX Video VAE | +| LTX Video 2 | Gemma3 | not specified | LTX Video 2 VAE | +| Lumina2 | Gemma2 | 2B | AutoencoderKL | +| Mage-Flow | Qwen3-VL | not specified | Mage-VAE | +| OmniGen | Integrated OmniGen encoder | not specified | AutoencoderKL | +| PixArt Sigma | T5 XXL v1.1 | 11B | AutoencoderKL | +| Qwen Image | Qwen2.5-VL | 7B | Qwen Image VAE | +| Sana | Gemma2 2B-IT | 2B | Sana AutoencoderDC | +| Sana Video | Gemma 2 | 2B | Wan VAE | +| SD 1.x/2.x (Legacy) | CLIP-L/14 | 123M | AutoencoderKL | +| Stable Diffusion 3 | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 | 123M + 694M + 11B | AutoencoderKL | +| Stable Diffusion XL | CLIP-L/14 + CLIP-G/14 | 123M + 694M | AutoencoderKL | +| Stable Cascade (Stage C) | CLIP-ViT-bigG-14 | 694M | Stable Cascade Stage C VAE | +| Wan Video | UMT5 | not specified | Wan VAE | +| Wan S2V | UMT5 | not specified | Wan VAE | +| Z-Image | Qwen3 4B | 4B | AutoencoderKL | +| Z-Image Omni | Qwen3 4B | 4B | AutoencoderKL | +| ZLab I1 | T5Gemma 2B | 2B | AutoencoderKL | + +
-| Modelo | Parâmetros | LoRA PEFT | Lycoris | Full-Rank | Quantização | Precisão mista | Checkpoint de gradiente | Flow Shift | TwinFlow | Self-Flow | LayerSync | Ref Inputs | ControlNet | Sliders† | Guia | -| --- | --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | --- | -| PixArt Sigma | 0.6B–0.9B | ✗ | ✓ | ✓ | int8 opcional | bf16 | ✓ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ | ✓ | [SIGMA.md](quickstart/SIGMA.md) | -| NVLabs Sana | 1.6B–4.8B | ✗ | ✓ | ✓ | int8 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [SANA.md](quickstart/SANA.md) | -| Kwai Kolors | 2.7B | ✓ | ✓ | ✓ | não recomendado | bf16 | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [KOLORS.md](quickstart/KOLORS.md) | -| Stable Diffusion 3 | 2B–8B | ✓ | ✓ | ✓ | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [SD3.md](quickstart/SD3.md) | -| Flux.1 | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [FLUX.md](quickstart/FLUX.md) | -| Flux.2 | 32B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [FLUX2.md](quickstart/FLUX2.md) | -| Flux Kontext | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✓ | ✓ | [FLUX_KONTEXT.md](quickstart/FLUX_KONTEXT.md) | -| Z-Image Turbo | 6B | ✓ | ✗ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZIMAGE.md](quickstart/ZIMAGE.md) | -| Krea2 | - | ✓ | ✗ | ✓* | int8 opcional | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✓ opt | ✗ | ✓ | [KREA2.md](quickstart/KREA2.pt-BR.md) | -| Boogu-Image 0.1 | - | ✓ | ✓ | ✓* | fp8 opcional | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ edit | ✗ | ✓ | [BOOGU_IMAGE.md](quickstart/BOOGU_IMAGE.pt-BR.md) | -| zlab i1 | 3B | ✓ | ✓ | ✓ | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZLAB_i1.md](quickstart/ZLAB_i1.pt-BR.md) | -| Ideogram 4 | 9B | ✓ | ✓ | ✓* | fp8 padrão, nf4 opcional | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [IDEOGRAM4.md](quickstart/IDEOGRAM4.pt-BR.md) | -| ACE-Step | 3.5B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✓ | [ACE_STEP.md](quickstart/ACE_STEP.md) | -| Chroma 1 | 8.9B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [CHROMA.md](quickstart/CHROMA.md) | -| Auraflow | 6B | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [AURAFLOW.md](quickstart/AURAFLOW.md) | -| HiDream I1 | 17B (8.5B MoE) | ✓ | ✓ | ✓* | int8/fp8/nf4 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [HIDREAM.md](quickstart/HIDREAM.md) | -| OmniGen | 3.8B | ✓ | ✓ | ✓ | int8/fp8 opcional | bf16 | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | [OMNIGEN.md](quickstart/OMNIGEN.md) | -| Stable Diffusion XL | 2.6B | ✓ | ✓ | ✓ | não recomendado | bf16 | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | [SDXL.md](quickstart/SDXL.md) | -| Lumina2 | 2B | ✓ | ✓ | ✓ | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | [LUMINA2.md](quickstart/LUMINA2.md) | -| Cosmos2 | 2B | ✓ | ✓ | ✓ | não recomendado | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [COSMOS2IMAGE.md](quickstart/COSMOS2IMAGE.md) | -| Cosmos3 | 16B-65B | ✓ | ✓ | ✓* | no_change primeiro | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | audio opt | ✗ | ✓ | [COSMOS3.md](quickstart/COSMOS3.pt-BR.md) | -| LTX Video | ~2.5B | ✓ | ✓ | ✓ | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [LTXVIDEO.md](quickstart/LTXVIDEO.md) | -| LTX Video 2 | 19B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LTXVIDEO2.md](quickstart/LTXVIDEO2.md) | -| Hunyuan Video 1.5 | 8.3B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [HUNYUANVIDEO.md](quickstart/HUNYUANVIDEO.md) | -| Wan 2.x | 1.3B–14B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN.md](quickstart/WAN.md) | -| Wan 2.2 S2V | 14B | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN_S2V.md](quickstart/WAN_S2V.md) | -| Qwen Image | 20B | ✓ | ✓ | ✓* | **obrigatório** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [QWEN_IMAGE.md](quickstart/QWEN_IMAGE.md) | -| Qwen Image Edit | 20B | ✓ | ✓ | ✓* | **obrigatório** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [QWEN_EDIT.md](quickstart/QWEN_EDIT.md) | -| Stable Cascade (C) | 1B, 3.6B prior | ✓ | ✓ | ✓* | não suportado | fp32 (obrigatório) | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [STABLE_CASCADE_C.md](quickstart/STABLE_CASCADE_C.md) | -| Kandinsky 5.0 Image | 6B (lite) | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ I2I | ✗ | ✓ | [KANDINSKY5_IMAGE.md](quickstart/KANDINSKY5_IMAGE.md) | -| Kandinsky 5.0 Video | 2B (lite), 19B (pro) | ✓ | ✓ | ✓* | int8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [KANDINSKY5_VIDEO.md](quickstart/KANDINSKY5_VIDEO.md) | -| LongCat-Video | 13.6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LONGCAT_VIDEO.md](quickstart/LONGCAT_VIDEO.md) | -| LongCat-Video Edit | 13.6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_VIDEO_EDIT.md](quickstart/LONGCAT_VIDEO_EDIT.md) | -| LongCat-Image | 6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [LONGCAT_IMAGE.md](quickstart/LONGCAT_IMAGE.md) | -| LongCat-Image Edit | 6B | ✓ | ✓ | ✓* | int8/fp8 opcional | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_EDIT.md](quickstart/LONGCAT_EDIT.md) | - -*✓ = suportado, ✓* = requer DeepSpeed/FSDP2 para full-rank, ✗ = não suportado, `✓+` indica que o checkpointing é recomendado devido à pressão de VRAM. Ref Inputs marca caminhos existentes de condicionamento por referência/edição/I2V; `opt` significa opcional e `req` significa obrigatório para o flavour de edição/I2V. TwinFlow ✓ significa suporte nativo quando `twinflow_enabled=true` (modelos de difusão precisam de `diff2flow_enabled+twinflow_allow_diff2flow`). Self-Flow ✓ significa suporte nativo para `crepa_enabled=true` com `crepa_feature_source=self_flow`, `use_ema=true` e `crepa_teacher_block_index` definido. LayerSync ✓ significa que o backbone expõe estados ocultos do transformer para autoalinhamento; ✗ marca backbones estilo UNet sem esse buffer. †Sliders se aplicam a LoRA e LyCORIS (incluindo LyCORIS full-rank “full”).* - -> ℹ️ O quickstart do Wan inclui presets das etapas 2.1 + 2.2 e o toggle de time-embedding. Flux Kontext cobre fluxos de edição construídos sobre o Flux.1. - -> ⚠️ Estes quickstarts são documentos vivos. Espere atualizações ocasionais conforme novos modelos chegam ou as receitas de treinamento melhoram. +*✓ = suportado, ✓* = suportado mas normalmente requer DeepSpeed/FSDP2 para treinamento full-rank, ✗ = não suportado. Ref Inputs marca rotas existentes de condicionamento por referência/edição/I2V; `opt` significa opcional e `req` significa obrigatório para o flavour de edição/I2V.* +*TwinFlow é nativo quando `twinflow_enabled=true`; modelos de difusão ainda exigem `diff2flow_enabled=true` e `twinflow_allow_diff2flow=true`. Self-Flow se refere ao suporte CREPA self-flow. LayerSync marca backbones que expõem hidden states para alinhamento.* ### Caminhos rápidos: Z-Image Turbo e Flux Schnell diff --git a/documentation/QUICKSTART.zh.md b/documentation/QUICKSTART.zh.md index 6b12ad619..303a25bea 100644 --- a/documentation/QUICKSTART.zh.md +++ b/documentation/QUICKSTART.zh.md @@ -2,55 +2,291 @@ **注意**：如需更高级的配置，请参阅[教程](TUTORIAL.md)和[选项参考](OPTIONS.md)。 +## 模型快速开始指南 + +| 模型 | 参数量 | 指南 | +| --- | --- | --- | +| ACE-Step | 3.5B | [ACE_STEP.zh.md](quickstart/ACE_STEP.zh.md) | +| Anima | Not specified | 暂无专用指南 | +| Auraflow | 6B | [AURAFLOW.zh.md](quickstart/AURAFLOW.zh.md) | +| Boogu-Image | Not specified | [BOOGU_IMAGE.zh.md](quickstart/BOOGU_IMAGE.zh.md) | +| Chroma 1 | 8.9B | [CHROMA.zh.md](quickstart/CHROMA.zh.md) | +| Cosmos2 | 2B-14B | [COSMOS2IMAGE.zh.md](quickstart/COSMOS2IMAGE.zh.md) | +| Cosmos3 | 16B-65B | [COSMOS3.zh.md](quickstart/COSMOS3.zh.md) | +| DeepFloyd IF | 0.4B-4.3B stages | 暂无专用指南 | +| ERNIE-Image | Not specified | [ERNIE.zh.md](quickstart/ERNIE.zh.md) | +| Flux.1 | 8B-12B | [FLUX.zh.md](quickstart/FLUX.zh.md)
[FLUX_KONTEXT.zh.md](quickstart/FLUX_KONTEXT.zh.md) | +| Flux.2 | 4B-32B | [FLUX2.zh.md](quickstart/FLUX2.zh.md) | +| HeartMuLa | 3B | [HEARTMULA.zh.md](quickstart/HEARTMULA.zh.md) | +| HiDream | 17B (8.5B MoE) | [HIDREAM.zh.md](quickstart/HIDREAM.zh.md) | +| Hunyuan Video | 8.3B | [HUNYUANVIDEO.zh.md](quickstart/HUNYUANVIDEO.zh.md) | +| Ideogram 4 | 9B | [IDEOGRAM4.zh.md](quickstart/IDEOGRAM4.zh.md) | +| Kandinsky 5.0 Image | 6B (lite) | [KANDINSKY5_IMAGE.zh.md](quickstart/KANDINSKY5_IMAGE.zh.md) | +| Kandinsky 5.0 Video | 2B lite, 19B pro | [KANDINSKY5_VIDEO.zh.md](quickstart/KANDINSKY5_VIDEO.zh.md) | +| Kwai Kolors | 2.7B | [KOLORS.zh.md](quickstart/KOLORS.zh.md) | +| Krea2 | Not specified | [KREA2.zh.md](quickstart/KREA2.zh.md) | +| LongCat Image | 6B | [LONGCAT_IMAGE.zh.md](quickstart/LONGCAT_IMAGE.zh.md)
[LONGCAT_EDIT.zh.md](quickstart/LONGCAT_EDIT.zh.md) | +| LongCat Video | 13.6B | [LONGCAT_VIDEO.zh.md](quickstart/LONGCAT_VIDEO.zh.md)
[LONGCAT_VIDEO_EDIT.zh.md](quickstart/LONGCAT_VIDEO_EDIT.zh.md) | +| LTX Video | ~2.5B | [LTXVIDEO.zh.md](quickstart/LTXVIDEO.zh.md) | +| LTX Video 2 | 19B | [LTXVIDEO2.zh.md](quickstart/LTXVIDEO2.zh.md) | +| Lumina2 | 2B | [LUMINA2.zh.md](quickstart/LUMINA2.zh.md) | +| Mage-Flow | 4B | [MAGEFLOW.zh.md](quickstart/MAGEFLOW.zh.md) | +| OmniGen | 3.8B | [OMNIGEN.zh.md](quickstart/OMNIGEN.zh.md) | +| PixArt Sigma | 0.6B-0.9B | [SIGMA.zh.md](quickstart/SIGMA.zh.md) | +| Qwen Image | 20B | [QWEN_IMAGE.zh.md](quickstart/QWEN_IMAGE.zh.md)
[QWEN_EDIT.zh.md](quickstart/QWEN_EDIT.zh.md) | +| Sana | 0.6B-4.8B | [SANA.zh.md](quickstart/SANA.zh.md) | +| Sana Video | 2B | [SANAVIDEO.zh.md](quickstart/SANAVIDEO.zh.md) | +| SD 1.x/2.x (Legacy) | 0.9B | 暂无专用指南 | +| Stable Diffusion 3 | 2B-8B | [SD3.zh.md](quickstart/SD3.zh.md) | +| Stable Diffusion XL | 3.5B | [SDXL.zh.md](quickstart/SDXL.zh.md) | +| Stable Cascade (Stage C) | 1B, 3.6B prior | [STABLE_CASCADE_C.zh.md](quickstart/STABLE_CASCADE_C.zh.md) | +| Wan Video | 1.3B-14B | [WAN.zh.md](quickstart/WAN.zh.md) | +| Wan S2V | 14B | [WAN_S2V.zh.md](quickstart/WAN_S2V.zh.md) | +| Z-Image | 6B | [ZIMAGE.zh.md](quickstart/ZIMAGE.zh.md) | +| Z-Image Omni | 6B | [ZIMAGE.zh.md](quickstart/ZIMAGE.zh.md) | +| ZLab I1 | 3B | [ZLAB_i1.zh.md](quickstart/ZLAB_i1.zh.md) | + ## 功能兼容性 -完整且最准确的功能矩阵，请参阅[主 README](https://github.com/bghira/SimpleTuner#model-architecture-support)。 +完整兼容性矩阵按功能领域拆分，以保持每张表易读。 -## 模型快速开始指南 +
+训练支持 + +| 模型 | PEFT LoRA | LyCORIS | 全秩 | ControlNet | Ref Inputs | +| --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓* | ✗ | ✗ | +| Anima | ✓ | ✓ | ✓* | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓* | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| Chroma 1 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓* | ✗ | audio opt | +| DeepFloyd IF | ✓ | ✓ | ✓ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Flux.1 | ✓ | ✓ | ✓* | ✓ | ✓ opt (Kontext) | +| Flux.2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| HeartMuLa | ✓ | ✓ | ✓* | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓* | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Ideogram 4 | ✓ | ✓ | ✓* | ✗ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓* | ✗ | ✓ I2I | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V | +| Kwai Kolors | ✓ | ✓ | ✓ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| LongCat Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| LongCat Video | ✓ | ✓ | ✓* | ✗ | ✓ opt/edit | +| LTX Video | ✓ | ✓ | ✓ | ✗ | ✓ I2V | +| LTX Video 2 | ✓ | ✓ | ✓* | ✗ | ✓ opt | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓* | ✗ | ✓ edit | +| OmniGen | ✓ | ✓ | ✓ | ✗ | ✗ | +| PixArt Sigma | ✗ | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓* | ✗ | ✓ req (Edit) | +| Sana | ✗ | ✓ | ✓ | ✗ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓* | ✓ | ✗ | +| Stable Diffusion XL | ✓ | ✓ | ✓ | ✓ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓* | ✗ | ✗ | +| Wan Video | ✓ | ✓ | ✓* | ✗ | ✓ I2V/VACE | +| Wan S2V | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image | ✓ | ✓ | ✓* | ✗ | ✗ | +| Z-Image Omni | ✓ | ✓ | ✓* | ✗ | ✓ opt (Edit) | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | ✗ | + +
+ +
+精度级别支持 + +| 模型 | 量化 | 混合精度 | +| --- | --- | --- | +| ACE-Step | int8 optional | bf16 | +| Anima | not specified | bf16 | +| Auraflow | int8/fp8/nf4 optional | bf16 | +| Boogu-Image | fp8 optional | bf16 | +| Chroma 1 | int8/fp8/nf4 optional | bf16 | +| Cosmos2 | int8 optional | bf16 | +| Cosmos3 | no_change first; int8 optional | bf16 | +| DeepFloyd IF | not recommended | bf16 | +| ERNIE-Image | int8 optional | bf16 | +| Flux.1 | int8/fp8/nf4 optional | bf16 | +| Flux.2 | int8/fp8/nf4 optional | bf16 | +| HeartMuLa | int8 optional | bf16 | +| HiDream | int8/fp8/nf4 optional | bf16 | +| Hunyuan Video | int8 optional | bf16 | +| Ideogram 4 | fp8 default, nf4 optional | bf16 | +| Kandinsky 5.0 Image | int8 optional | bf16 | +| Kandinsky 5.0 Video | int8 optional | bf16 | +| Kwai Kolors | not recommended | bf16 | +| Krea2 | int8 optional | bf16 | +| LongCat Image | int8/fp8 optional | bf16 | +| LongCat Video | int8/fp8 optional | bf16 | +| LTX Video | int8/fp8 optional | bf16 | +| LTX Video 2 | int8/fp8 optional | bf16 | +| Lumina2 | int8 optional | bf16 | +| Mage-Flow | fp8 optional | bf16 | +| OmniGen | int8/fp8 optional | bf16 | +| PixArt Sigma | int8 optional | bf16 | +| Qwen Image | required (int8/nf4) | bf16 | +| Sana | int8 optional | bf16 | +| Sana Video | not recommended for full | bf16 | +| SD 1.x/2.x (Legacy) | int8/nf4 optional | bf16 | +| Stable Diffusion 3 | int8/fp8/nf4 optional | bf16 | +| Stable Diffusion XL | int8/nf4 optional | bf16 | +| Stable Cascade (Stage C) | not supported | fp32 required | +| Wan Video | int8 optional | bf16 | +| Wan S2V | int8 optional | bf16 | +| Z-Image | int8 optional | bf16 | +| Z-Image Omni | int8 optional | bf16 | +| ZLab I1 | int8 optional | bf16 | + +
+ +
+检查点粒度 + +| 模型 | Gradient Checkpoint | Interval | Segment Stride | Attention Offload | +| --- | :---: | :---: | :---: | :---: | +| ACE-Step | ✓ | ✓ | ✓ | ✗ | +| Anima | ✓ | ✗ | ✗ | ✗ | +| Auraflow | ✓ | ✓ | ✓ | ✗ | +| Boogu-Image | ✓ | ✓ | ✓ | ✗ | +| Chroma 1 | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | ✓ | ✓ | ✓ | ✗ | +| Cosmos3 | ✓ | ✓ | ✓ | ✗ | +| DeepFloyd IF | ✓ | ✗ | ✗ | ✗ | +| ERNIE-Image | ✓ | ✓ | ✓ | ✗ | +| Flux.1 | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | ✓ | ✗ | ✗ | ✗ | +| HiDream | ✓ | ✓ | ✓ | ✗ | +| Hunyuan Video | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | ✓ | ✓ | ✓ | ✗ | +| Kandinsky 5.0 Image | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | ✓ | ✗ | ✗ | ✗ | +| Krea2 | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | ✓ | ✓ | ✓ | ✓ | +| LTX Video | ✓ | ✓ | ✓ | ✗ | +| LTX Video 2 | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | ✓ | ✓ | ✓ | ✗ | +| Mage-Flow | ✓ | ✓ | ✓ | ✓ | +| OmniGen | ✓ | ✗ | ✗ | ✗ | +| PixArt Sigma | ✓ | ✓ | ✓ | ✗ | +| Qwen Image | ✓ | ✓ | ✓ | ✗ | +| Sana | ✓ | ✓ | ✓ | ✗ | +| Sana Video | ✓ | ✓ | ✓ | ✗ | +| SD 1.x/2.x (Legacy) | ✓ | ✗ | ✗ | ✗ | +| Stable Diffusion 3 | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | ✓ | ✗ | ✗ | ✗ | +| Stable Cascade (Stage C) | ✓ | ✓ | ✓ | ✗ | +| Wan Video | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | ✓ | ✓ | ✓ | ✗ | +| Z-Image | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | ✓ | ✗ | ✗ | ✗ | +| ZLab I1 | ✓ | ✓ | ✓ | ✗ | + +
+ +
+Flow、蒸馏与对齐 + +| 模型 | Prediction | Flow Shift | TwinFlow | Self-Flow | LayerSync | Sliders | +| --- | --- | :---: | :---: | :---: | :---: | :---: | +| ACE-Step | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Anima | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Auraflow | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Boogu-Image | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Chroma 1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Cosmos2 | sample | ✗ | ✗ | ✓ | ✓ | ✓ | +| Cosmos3 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| DeepFloyd IF | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| ERNIE-Image | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| Flux.1 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Flux.2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| HeartMuLa | autoregressive next-token | ✗ | ✗ | ✗ | ✗ | ✗ | +| HiDream | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Hunyuan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Ideogram 4 | flow matching | ✓ | ✗ | ✗ | ✗ | ✓ | +| Kandinsky 5.0 Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kandinsky 5.0 Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Kwai Kolors | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Krea2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LongCat Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| LTX Video 2 | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Lumina2 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Mage-Flow | flow matching | ✓ | ✓ | ✗ | ✓ | ✓ | +| OmniGen | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| PixArt Sigma | epsilon | ✗ | ✗ | ✓ | ✓ | ✓ | +| Qwen Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Sana Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| SD 1.x/2.x (Legacy) | epsilon / v-pred | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Diffusion 3 | flow matching | ✓ (SLG) | ✓ | ✓ | ✓ | ✓ | +| Stable Diffusion XL | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Stable Cascade (Stage C) | epsilon | ✗ | ✗ | ✗ | ✗ | ✓ | +| Wan Video | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Wan S2V | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | +| Z-Image | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| Z-Image Omni | flow matching | ✓ | ✓ | ✓ | ✓ | ✓ | +| ZLab I1 | flow matching | ✓ | ✗ | ✓ | ✓ | ✓ | + +
+ +
+文本编码器与 VAE 类型 + +| 模型 | Text Encoders | Text Encoder Params | VAE | +| --- | --- | --- | --- | +| ACE-Step | UMT5 Encoder | 0.6B | Music DCAE | +| Anima | Qwen3 0.6B | 0.6B | Qwen Image VAE | +| Auraflow | Pile T5 | not specified | AutoencoderKL | +| Boogu-Image | Qwen3-VL | not specified | AutoencoderKL | +| Chroma 1 | T5 XXL v1.1 | 11B | AutoencoderKL | +| Cosmos2 | T5 11B | 11B | Wan VAE | +| Cosmos3 | Cosmos3 reasoner | not specified | Wan/Cosmos VAE | +| DeepFloyd IF | T5 XXL v1.1 | 11B | None | +| ERNIE-Image | ERNIE text encoder | not specified | Flux.2 VAE | +| Flux.1 | CLIP-L/14 + T5 XXL v1.1 | 123M + 11B | AutoencoderKL | +| Flux.2 | Mistral-Small-3.1-24B | 24B | Flux.2 VAE | +| HeartMuLa | None | N/A | HeartCodec tokens | +| HiDream | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 + Llama | 123M + 694M + 11B + not specified | AutoencoderKL | +| Hunyuan Video | Hunyuan LLM | not specified | Hunyuan Video 3D VAE | +| Ideogram 4 | Qwen3-VL-8B-Instruct | 8B | Ideogram AutoEncoder | +| Kandinsky 5.0 Image | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Flux VAE (AutoencoderKL) | +| Kandinsky 5.0 Video | Qwen2.5-VL + CLIP-L/14 | 7B + 123M | Hunyuan Video VAE | +| Kwai Kolors | ChatGLM-6B | 6B | AutoencoderKL | +| Krea2 | Qwen3VL | not specified | Qwen Image VAE | +| LongCat Image | Qwen2.5-VL | 7B | AutoencoderKL | +| LongCat Video | Qwen2.5-VL | 7B | Wan VAE | +| LTX Video | T5 XXL v1.1 | 11B | LTX Video VAE | +| LTX Video 2 | Gemma3 | not specified | LTX Video 2 VAE | +| Lumina2 | Gemma2 | 2B | AutoencoderKL | +| Mage-Flow | Qwen3-VL | not specified | Mage-VAE | +| OmniGen | Integrated OmniGen encoder | not specified | AutoencoderKL | +| PixArt Sigma | T5 XXL v1.1 | 11B | AutoencoderKL | +| Qwen Image | Qwen2.5-VL | 7B | Qwen Image VAE | +| Sana | Gemma2 2B-IT | 2B | Sana AutoencoderDC | +| Sana Video | Gemma 2 | 2B | Wan VAE | +| SD 1.x/2.x (Legacy) | CLIP-L/14 | 123M | AutoencoderKL | +| Stable Diffusion 3 | CLIP-L/14 + CLIP-G/14 + T5 XXL v1.1 | 123M + 694M + 11B | AutoencoderKL | +| Stable Diffusion XL | CLIP-L/14 + CLIP-G/14 | 123M + 694M | AutoencoderKL | +| Stable Cascade (Stage C) | CLIP-ViT-bigG-14 | 694M | Stable Cascade Stage C VAE | +| Wan Video | UMT5 | not specified | Wan VAE | +| Wan S2V | UMT5 | not specified | Wan VAE | +| Z-Image | Qwen3 4B | 4B | AutoencoderKL | +| Z-Image Omni | Qwen3 4B | 4B | AutoencoderKL | +| ZLab I1 | T5Gemma 2B | 2B | AutoencoderKL | + +
-| 模型 | 参数量 | PEFT LoRA | Lycoris | 全秩 | 量化 | 混合精度 | 梯度检查点 | Flow Shift | TwinFlow | Self-Flow | LayerSync | Ref Inputs | ControlNet | Sliders† | 指南 | -| --- | --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | --- | -| PixArt Sigma | 0.6B–0.9B | ✗ | ✓ | ✓ | int8 可选 | bf16 | ✓ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ | ✓ | [SIGMA.md](quickstart/SIGMA.md) | -| NVLabs Sana | 1.6B–4.8B | ✗ | ✓ | ✓ | int8 可选 | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [SANA.md](quickstart/SANA.md) | -| Kwai Kolors | 2.7B | ✓ | ✓ | ✓ | 不推荐 | bf16 | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [KOLORS.md](quickstart/KOLORS.md) | -| Stable Diffusion 3 | 2B–8B | ✓ | ✓ | ✓ | int8/fp8/nf4 可选 | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [SD3.md](quickstart/SD3.md) | -| Flux.1 | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 可选 | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [FLUX.md](quickstart/FLUX.md) | -| Flux.2 | 32B | ✓ | ✓ | ✓* | int8/fp8/nf4 可选 | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [FLUX2.md](quickstart/FLUX2.md) | -| Flux Kontext | 8B–12B | ✓ | ✓ | ✓* | int8/fp8/nf4 可选 | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✓ | ✓ | [FLUX_KONTEXT.md](quickstart/FLUX_KONTEXT.md) | -| Z-Image Turbo | 6B | ✓ | ✗ | ✓* | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZIMAGE.md](quickstart/ZIMAGE.md) | -| Krea2 | - | ✓ | ✗ | ✓* | int8 可选 | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✓ opt | ✗ | ✓ | [KREA2.md](quickstart/KREA2.zh.md) | -| Boogu-Image 0.1 | - | ✓ | ✓ | ✓* | fp8 可选 | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ edit | ✗ | ✓ | [BOOGU_IMAGE.md](quickstart/BOOGU_IMAGE.zh.md) | -| zlab i1 | 3B | ✓ | ✓ | ✓ | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [ZLAB_i1.md](quickstart/ZLAB_i1.zh.md) | -| Ideogram 4 | 9B | ✓ | ✓ | ✓* | fp8 默认，nf4 可选 | bf16 | ✓+ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [IDEOGRAM4.md](quickstart/IDEOGRAM4.zh.md) | -| ACE-Step | 3.5B | ✓ | ✓ | ✓* | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✓ | [ACE_STEP.md](quickstart/ACE_STEP.md) | -| Chroma 1 | 8.9B | ✓ | ✓ | ✓* | int8/fp8/nf4 可选 | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [CHROMA.md](quickstart/CHROMA.md) | -| Auraflow | 6B | ✓ | ✓ | ✓* | int8/fp8/nf4 可选 | bf16 | ✓+ | ✓ (SLG) | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [AURAFLOW.md](quickstart/AURAFLOW.md) | -| HiDream I1 | 17B (8.5B MoE) | ✓ | ✓ | ✓* | int8/fp8/nf4 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | [HIDREAM.md](quickstart/HIDREAM.md) | -| OmniGen | 3.8B | ✓ | ✓ | ✓ | int8/fp8 可选 | bf16 | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | [OMNIGEN.md](quickstart/OMNIGEN.md) | -| Stable Diffusion XL | 2.6B | ✓ | ✓ | ✓ | 不推荐 | bf16 | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | [SDXL.md](quickstart/SDXL.md) | -| Lumina2 | 2B | ✓ | ✓ | ✓ | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | [LUMINA2.md](quickstart/LUMINA2.md) | -| Cosmos2 | 2B | ✓ | ✓ | ✓ | 不推荐 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [COSMOS2IMAGE.md](quickstart/COSMOS2IMAGE.md) | -| Cosmos3 | 16B-65B | ✓ | ✓ | ✓* | no_change first | bf16 | ✓ | ✓ | ✗ | ✗ | ✗ | audio opt | ✗ | ✓ | [COSMOS3.md](quickstart/COSMOS3.zh.md) | -| LTX Video | ~2.5B | ✓ | ✓ | ✓ | int8/fp8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [LTXVIDEO.md](quickstart/LTXVIDEO.md) | -| LTX Video 2 | 19B | ✓ | ✓ | ✓* | int8/fp8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LTXVIDEO2.md](quickstart/LTXVIDEO2.md) | -| Hunyuan Video 1.5 | 8.3B | ✓ | ✓ | ✓* | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [HUNYUANVIDEO.md](quickstart/HUNYUANVIDEO.md) | -| Wan 2.x | 1.3B–14B | ✓ | ✓ | ✓* | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN.md](quickstart/WAN.md) | -| Wan 2.2 S2V | 14B | ✓ | ✓ | ✓* | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [WAN_S2V.md](quickstart/WAN_S2V.md) | -| Qwen Image | 20B | ✓ | ✓ | ✓* | **必需** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [QWEN_IMAGE.md](quickstart/QWEN_IMAGE.md) | -| Qwen Image Edit | 20B | ✓ | ✓ | ✓* | **必需** (int8/nf4) | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [QWEN_EDIT.md](quickstart/QWEN_EDIT.md) | -| Stable Cascade (C) | 1B, 3.6B prior | ✓ | ✓ | ✓* | 不支持 | fp32 (必需) | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | [STABLE_CASCADE_C.md](quickstart/STABLE_CASCADE_C.md) | -| Kandinsky 5.0 Image | 6B (lite) | ✓ | ✓ | ✓* | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ I2I | ✗ | ✓ | [KANDINSKY5_IMAGE.md](quickstart/KANDINSKY5_IMAGE.md) | -| Kandinsky 5.0 Video | 2B (lite), 19B (pro) | ✓ | ✓ | ✓* | int8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ I2V | ✗ | ✓ | [KANDINSKY5_VIDEO.md](quickstart/KANDINSKY5_VIDEO.md) | -| LongCat-Video | 13.6B | ✓ | ✓ | ✓* | int8/fp8 可选 | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ opt | ✗ | ✓ | [LONGCAT_VIDEO.md](quickstart/LONGCAT_VIDEO.md) | -| LongCat-Video Edit | 13.6B | ✓ | ✓ | ✓* | int8/fp8 可选 | bf16 | ✓+ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_VIDEO_EDIT.md](quickstart/LONGCAT_VIDEO_EDIT.md) | -| LongCat-Image | 6B | ✓ | ✓ | ✓* | int8/fp8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✓ | [LONGCAT_IMAGE.md](quickstart/LONGCAT_IMAGE.md) | -| LongCat-Image Edit | 6B | ✓ | ✓ | ✓* | int8/fp8 可选 | bf16 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ req | ✗ | ✓ | [LONGCAT_EDIT.md](quickstart/LONGCAT_EDIT.md) | - -*✓ = 支持，✓* = 全秩训练需要 DeepSpeed/FSDP2，✗ = 不支持，`✓+` 表示由于 VRAM 压力建议启用检查点。Ref Inputs 仅标记现有参考/编辑/I2V 条件路径；`opt` 表示可选，`req` 表示该编辑/I2V flavour 需要它。TwinFlow ✓ 表示当 `twinflow_enabled=true` 时原生支持（扩散模型需要 `diff2flow_enabled+twinflow_allow_diff2flow`）。Self-Flow ✓ 表示原生支持 `crepa_enabled=true`、`crepa_feature_source=self_flow`、`use_ema=true` 且设置 `crepa_teacher_block_index`。LayerSync ✓ 表示骨干网络暴露 transformer 隐藏状态用于自对齐；✗ 标记没有该缓冲区的 UNet 风格骨干网络。†Sliders 适用于 LoRA 和 LyCORIS（包括全秩 LyCORIS "full"）。* - -> 注：Wan 快速开始包含 2.1 + 2.2 阶段预设和时间嵌入切换。Flux Kontext 涵盖基于 Flux.1 构建的编辑工作流程。 - -> 警告：这些快速开始指南是持续更新的文档。随着新模型的发布或训练方案的改进，预计会有不定期更新。 +*✓ = 支持，✓* = 支持但 full-rank training 通常需要 DeepSpeed/FSDP2，✗ = 不支持。Ref Inputs 标记现有 reference/edit/I2V conditioning paths；`opt` 表示可选，`req` 表示 edit/I2V flavour 必需。* +*TwinFlow 在 `twinflow_enabled=true` 时为原生支持；diffusion models 仍需要 `diff2flow_enabled=true` 和 `twinflow_allow_diff2flow=true`。Self-Flow 指 CREPA self-flow support。LayerSync 标记公开 hidden states 用于 alignment 的 backbones。* ### 快速通道：Z-Image Turbo 和 Flux Schnell