d0gr/sdxl-lightning-onnx-webgpu

SDXL-Lightning ONNX (WebGPU, fp16)

ONNX build of ByteDance/SDXL-Lightning (4-step) for in-browser inference via onnxruntime-web (WebGPU). Includes the UNet, both text encoders, VAE and tokenizers. UNet is re-sharded into <2 GB parts.

Used by the Generate AI Images extension (local SDXL generation in the browser, no server).

• Resolution: 1024×1024
• Steps: 4–6, guidance ~1.5
• I/O: fp32 (fp16 internals)
• Size: ~6.9 GB

License

Derivative work combining three upstream sources (all permissive; commercial use permitted, subject to the RAIL++-M use restrictions):

• UNet — ByteDance/SDXL-Lightning — CreativeML Open RAIL++-M
• SDXL base (text encoders) — stabilityai/stable-diffusion-xl-base-1.0 — CreativeML Open RAIL++-M
• VAE — madebyollin/sdxl-vae-fp16-fix — MIT

The combined work is released under CreativeML Open RAIL++-M (the most restrictive of the three).