mlc-ai/web-stable-diffusion

Leverages Apache TVM Unity's machine learning compilation to transform Hugging Face Stable Diffusion models through TorchDynamo/Torch FX capture, then automatically optimizes them using TensorIR and MetaSchedule for WebGPU execution. The workflow enables Python-first model development that compiles to WebAssembly and JavaScript runtimes, with careful memory management to fit large diffusion models within browser constraints. Integrates with PyTorch, Hugging Face diffusers, Rust tokenizers, and WebGPU for hardware-accelerated inference across diverse client environments.

3,714 stars and 17 monthly downloads. No commits in the last 6 months. Available on PyPI.