Generate AI images and videos — on your own GPU or in the cloud. One app.
Free · No subscription · Windows & macOS
English · 简体中文
Download · Quick start · Website · Contributing · Security
| Platform | Download |
|---|---|
| Windows 10/11 (x64) | Installer · Portable .exe |
| macOS 12+ (Intel & Apple Silicon) | .dmg |
| Linux | No prebuilt binary yet — build from source |
All versions and checksums.txt (SHA-256) are on the
Releases page.
- Download the installer for your OS above and run it.
- Open the app — it installs its own inference engine and detects your GPU. Nothing touches your system Python.
- Pick a model, hit Generate — the weights download automatically and every model ships pre-tuned, so your first image already looks right.
No GPU? Switch the toggle to Cloud in step 3 and generate on imference.com instead.
The app isn't code-signed yet, so your OS shows a one-time warning. This is expected — here's how to get past it:
- Windows — SmartScreen popup → click More info → Run anyway.
- macOS — right-click the app → Open → Open.
If macOS still refuses:
xattr -cr "/Applications/Imference Desktop.app"
You can verify your download against checksums.txt from the release.
The app checks for new versions at startup and shows a banner when one is
available; code signing and silent in-app auto-update are on the roadmap.
- ⚡ One-click setup — the app installs an isolated inference engine for you (nothing touches your system Python), detects your GPU, and starts and stops the engine on demand. No terminal, no CUDA wrestling.
- 🖥️ Local generation, $0 per image — run eight model families on your own GPU: SDXL, SD 1.5, Z-Image, FLUX, Chroma, Qwen-Image, Anima and Krea 2. Your prompts and images never leave your machine.
- 🧩 Preconfigured models, zero setup — pick a model from the curated catalog and hit Generate: the weights download automatically, and every model ships pre-tuned (steps, CFG, resolutions, quality tags, negative prompt) so your first image already looks right.
- 📦 Bring your own model — load any local
.safetensorscheckpoint (e.g. downloaded from Civitai) and pick its family. The file is used in place — nothing is copied or uploaded. - ☁️ Cloud when you want it — no GPU or a weak one? Generate images and video on imference.com from the same interface. Pay with an API key (credits) or x402 (USDC on Base) — the x402 route needs no account at all.
- 🎛️ Real controls, simple interface — format, steps, CFG, seed, quality tags, negative prompt, image-to-image. Stack generations in a queue, and even switch models mid-run — the app finishes the current image first.
- 🗂️ A gallery that remembers everything — every image and video is saved with its prompt, model and settings. Filter, search, and review in a fullscreen viewer.
| Mode | What you need |
|---|---|
| Local | Windows: an NVIDIA GPU (CUDA), or an AMD Radeon RX 7000/9000 (ROCm preview — needs Python 3.12 + a recent Adrenalin driver) · macOS: Apple Silicon. ~6–13 GB disk per model. |
| Cloud | Any machine — an imference.com API key, or a funded x402 wallet. |
Linux (NVIDIA CUDA / AMD ROCm) is supported by the engine and builds from source, but there's no prebuilt binary yet — see Build from source.
Learn more about the app on the imference.com/desktop page.
This is an early public release — rough edges are expected.
- 🐛 Something broken? Open an issue — it directly shapes what gets built next.
- 💬 A question, or not sure it's a bug? Discussions → Q&A — GPU support, install trouble, how-to.
- 🖼️ Made something good? Show and tell.
Build from source
Built with Wails v3 (Go + React). This is also how to run it on Linux, which the release workflow doesn't package yet.
Requires Go 1.25+, Node 20+, and the Wails3 CLI:
go install github.com/wailsapp/wails/v3/cmd/wails3@v3.0.0-alpha2.116
wails3 dev # run in dev mode (hot reload)
wails3 package # production build + platform packaging (NSIS on Windows)Releases are built automatically by GitHub Actions on a pushed v* tag
(see .github/workflows/release.yml).
See CONTRIBUTING.md for the repo layout, the checks to run before a PR, and how to add a language.
MIT © Publikey Sàrl
Model weights are not covered by this license — each model on the catalog carries its own terms (CreativeML, FLUX non-commercial, Apache-2.0…). Check the license of any model you use, especially commercially.
Found a security issue? See SECURITY.md — please don't open a public issue.