Rust-native inference
Candle runs supported diffusion pipelines without a Python runtime hiding under the floorboards.
mold
Local image generation. No cloud, no Python, no fuss.
Generate images and short video clips on your own GPU with a fast Rust CLI, web studio, desktop app, and remote rendering support for modern diffusion models.
- Language
- Rust
- License
- MIT
- Stars
- 34
- Forks
- 0
What it fixes
Local diffusion usually starts with Python archaeology and ends with your GPU judging you. mold ships the runtime, downloads models, and gets out of the way.
mold is the shortest path from a prompt to pixels on hardware you control.
- Python environments held together by ancient wheel metadata
- Model setup spread across scripts and mystery directories
- Cloud APIs charging rent for hardware already under your desk
- Local and remote GPUs behaving like separate products
What it actually does
A useful product surface, not another README wearing a blazer.
Modern model support
Generate with FLUX, SD 1.5, SDXL, Z-Image, and supported image-to-video models.
CLI and terminal UI
Generate, preview, pipe, manage models, and inspect jobs without leaving the terminal.
Mold Studio
Use the bundled web interface for creation, history, models, machines, and settings.
Desktop and iPhone
Drive the same generation system from native desktop and mobile clients.
Remote and cloud GPUs
Route work to remembered machines or provision RunPod capacity when the local GPU taps out.
Pick your poison
Run it the way that annoys you least.
Install
curl -fsSL https://raw.githubusercontent.com/utensils/mold/main/install.sh | sh
Generate
mold run "a cat riding a motorcycle through neon-lit streets"
Nix
nix run github:utensils/mold -- run "a cat"
Plays well with the tools already causing trouble
- FLUX
- Stable Diffusion
- SDXL
- Z-Image
- Candle
- CUDA
- Metal
- RunPod
Under the hood
Simple enough to explain without a thirty-slide architecture deck.
- 01
Prompt and model
Resolve a local model variant and generation settings from CLI or Studio.
- 02
Candle runtime
Rust loads the pipeline and runs inference through CUDA, Metal, or CPU backends.
- 03
Machine routing
Keep the job local or send it to a remembered server or cloud GPU.
- 04
Durable library
Outputs, metadata, models, and job state stay available across every client surface.
Make the GPU earn its electricity
Install one binary, pull a model, and start generating without summoning a Python environment.
Latest release: v0.20.2