tutorial
Porting the Moebius 0.2B image inpainting model to run in the browser with Claude Code
For builders, this means lower-cost deployment of inpainting functionality (no GPU servers) and a pattern for leveraging coding agents to accelerate model porting and integration.
What happened
Simon Willison, creator of Datasette, ported the Moebius 0.2B image inpainting model to run entirely in the browser using WebGPU. Originally requiring PyTorch and NVIDIA CUDA, the lightweight model (0.2B parameters) can now remove and fill image regions client-side. Willison accomplished this as a side project while waiting for AI coding agents to complete larger tasks—specifically using Claude Code to handle the port. The demo is live on his GitHub page. This showcases both the feasibility of running capable AI models in the browser without server-side GPUs and the productivity boost from using coding agents to parallelize development work.
Key takeaways
- Moebius is a 0.2B parameter image inpainting model originally dependent on PyTorch and CUDA.
- Simon Willison ported the model to WebGPU for browser execution using Claude Code.
- The porting was done as a side project while waiting for an AI agent to finish a Datasette feature.
- The tool allows users to upload images, mark areas to remove, and run inpainting client-side.
- The approach demonstrates running small but effective models in the browser without server-side infrastructure.
Why it matters
For builders, this means lower-cost deployment of inpainting functionality (no GPU servers) and a pattern for leveraging coding agents to accelerate model porting and integration.
This is an original editorial digest by AI Workflow Center. Full reporting at the source:
Read the original on Simon WillisonMore AI news
All news →
Run Your Own AI Directory