diffusers.js WebGPU demohttps://islamov.ai/stable-diffusion-webgpu/

diffusers.js WebGPU demo

šŸš€ Check out diffusers.js WebGPU demo! šŸ¤Æ Built with cutting-edge tech: āœØ Supports FP16 with Chrome Canary 121+ šŸŒŸ Utilizes PNDM Scheduler & VAE šŸ§¬ Control image similarity with Guidance Scale šŸ› ļø Detailed troubleshooting steps included! Follow @dakenf on GitHub for more! #WebGPU #AI #Innovation šŸ¤–šŸ’»

  • Use latest Chrome with "Experimental WebAssembly" flag enabled and "Experimental WebAssembly JavaScript Promise Integration (JSPI)" flag enabled.
  • Require Chrome Canary 119 or newer for optimal performance.
  • Chrome Canary 121 or higher needed for FP16 support.
  • Built with diffusers.js for specific functionality.
  • Utilize Prompt Negative Prompt Number of inference steps (PNDM Scheduler: i+1 rule).
  • Adjust Guidance Scale to control similarity of generated image to prompt.
  • Utilize Seed to create initial random noise for model.
  • Option to run VAE after each step in the pipeline.
  • Download model to browser cache to access editable settings.
  • Follow troubleshooting steps for protobuf parsing and memory errors.
  • Implementation involved porting StableDiffusionPipeline to JS, patching onnxruntime and emscripten+binaryen for >4GB memory allocation support, and compliance with WebAssembly spec and V8 engine.