Stable Diffusion Webgpu

Stable Diffusion WebGPU: An AI Tool for Image Generation

Are you looking for a powerful AI tool that allows you to generate high-quality images? Look no further than Stable Diffusion WebGPU. This web-based application, built on the create-react-app framework, offers a user-friendly interface and impressive capabilities for image creation.

Key Features:

• Easy-to-use interface for loading models, running the image generation process, and viewing results
• Cached model files: no repeated downloads required
• Optimized CPU performance for accurate and high-performing image generation
• FAQ section for troubleshooting guidance
• Open-source code available on GitHub for local usage and customization

When using Stable Diffusion WebGPU, it is crucial to have JavaScript enabled and the latest version of Chrome. To unlock the application's full potential, make sure to enable the "Experimental WebAssembly" and "Experimental WebAssembly JavaScript Promise Integration (JSPI)" flags in your browser settings.

The image generation process consists of a series of inference steps, with each step taking approximately 1 minute and an additional 10 seconds for the VAE decoder to generate the image. For acceptable results, we recommend a minimum of 20 steps. However, for a quick demonstration, 3 steps will suffice.

Please note that keeping the DevTools open while using the application may slow down the process by approximately 2 times.

Use Cases:

• Artists and designers who want to explore new creative possibilities and generate unique images
• Researchers and scientists working with image-based datasets
• Developers interested in learning about image generation algorithms

While Stable Diffusion WebGPU leverages the power of GPU, the implementation in onnxruntime is still in its early stage. Some operations may be incomplete, leading to data continuously transferred between the CPU and GPU, affecting performance. The tool does not support multi-threading, and 64-bit memory creation with SharedArrayBuffer is limited due to WebAssembly constraints.

The developer is actively working on improving the tool by proposing specification changes and engine patches. Additionally, a patched version of onnxruntime is provided for using large language models with transformers.js, although reliability in all scenarios is not guaranteed. The developer also plans to submit a pull request to the onnxruntime repository.

Give Stable Diffusion WebGPU a try today and unlock the full potential of image generation!