Skip to content

DEPENDENCIES.md

Latest commit

 

History

History
132 lines (94 loc) · 4.33 KB

DEPENDENCIES.md

File metadata and controls

132 lines (94 loc) · 4.33 KB

External Dependencies

All external dependencies required to run the Medium Format Studio web UI.

Runtime Dependencies

Node.js

  • Version: 18 or higher (developed on v25; see .nvmrc)
  • Install: https://nodejs.org/ or via nvm
  • Check: node --version && npm --version

Python

ComfyUI

  • Repository: https://github.com/comfyanonymous/ComfyUI
  • Install: 
    git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txt
  • Start (CORS required): 
    python main.py --enable-cors-header

ComfyUI Custom Nodes

The Medium Format Studio workflow uses several custom nodes. Install each via ComfyUI Manager or by cloning into ComfyUI/custom_nodes/.

Node Pack Required For
comfyui-kjnodes EmptyLatentImageCustomPresets (film format presets)
rgthree-comfy Power Lora Loader, Seed (rgthree)
ComfyUI-SeedVR2 SeedVR2 AI upscaling (Final Print stage)

Installing via ComfyUI Manager (recommended):

  1. Install ComfyUI Manager: cd ComfyUI/custom_nodes && git clone https://github.com/ltdrdata/ComfyUI-Manager.git
  2. Restart ComfyUI
  3. Open ComfyUI web UI → Manager → Install Missing Custom Nodes

Model Files

Place models in the indicated directories within your ComfyUI installation. All are required unless noted.

Diffusion Model

Place in ComfyUI/models/unet/ or ComfyUI/models/diffusion_models/:

  • flux-2-klein-9b.safetensors (FP16, ~18GB) — primary model
  • or flux-2-klein-9b-Q8_0.gguf (~9GB), flux-2-klein-9b-Q6_K.gguf (~7GB), etc.

The UI auto-discovers all flux-2-klein-9b* variants present on the server and lets you select among them. At least one variant is required.

Text Encoder

Place in ComfyUI/models/clip/:

  • qwen_3_8b_fp8mixed.safetensors (~8GB)

VAE

Place in ComfyUI/models/vae/:

  • flux2-vae.safetensors (~300MB)

AI Upscaler (Final Print stage)

Place in ComfyUI/models/ — check SeedVR2 docs for exact subdirectory:

  • seedvr2_ema_7b_sharp_fp16.safetensors (~14GB)

LoRAs

Place in ComfyUI/models/loras/FluxKlein/:

  • detail_slider_klein_9b_20260123_065513.safetensors
  • klein_slider_chiaroscuro.safetensors

LoRAs are optional at runtime (each can be disabled in the UI), but both files must be present for the workflow to load without errors.

Disk Space

Component Approximate Size
node_modules ~200MB
ComfyUI installation ~500MB
Flux 2 Klein 9B (FP16 .safetensors) ~18GB
— or GGUF variants 7–9GB
Qwen 8B text encoder ~8GB
Flux2 VAE ~300MB
SeedVR2 7B upscaler ~14GB
LoRAs (both) ~1GB

Total (FP16 path): ~42GB Total (GGUF Q8 path): ~33GB

GPU / Compute Requirements

  • NVIDIA GPU (CUDA): Recommended; supports FlashAttention, SageAttention, full performance
  • AMD GPU (ROCm): Supported by ComfyUI; performance varies
  • Apple Silicon (MPS): Works via ComfyUI's MPS backend; lacks CUDA-only optimizations (FlashAttention, Triton). GGUF quantized variants reduce VRAM requirements.
  • CPU: Technically supported but impractically slow for these model sizes

VRAM Guidance

Setup Minimum VRAM
Klein 9B FP16 ~20GB
Klein 9B Q8 GGUF ~12GB
Klein 9B Q6_K GGUF ~10GB
+ SeedVR2 (Final Print) +8–10GB (loaded separately)

ComfyUI offloads models as needed; lower VRAM is possible with slower generation.

Network / Ports

  • 8188 — ComfyUI backend (default; adjustable in ComfyUI settings)
  • 5173 — Vite dev server

Both ports must be free on the local machine. The web UI can also connect to a remote ComfyUI instance (e.g. RunPod) — configure the URL via the server settings panel in the UI.

Updating

# Web UI
git pull && npm install

# ComfyUI
cd /path/to/ComfyUI && git pull && pip install -r requirements.txt

# Custom nodes (via ComfyUI Manager)
# Open ComfyUI web UI → Manager → Update All