Use any model with diffusers without waiting for an official integration
Blog post from Hugging Face
A guide by Alvaro Somoza explains how to use a model that is not officially supported by Diffusers by integrating it as custom blocks in Modular Diffusers, with an AI coding agent helping research, convert weights, build a pipeline, and test the result. Using Idol, which turns a single image into 3D Gaussian splats and camera estimates, the author shows how the integration can avoid the model’s complex, platform-specific dependencies and work alongside other models in a shared environment, while supporting Diffusers optimizations. The process includes reviewing the agent’s plan, comparing each component against the original implementation in full precision, testing the complete pipeline, and visually inspecting outputs; the Idol port passed these checks after several iterations. The author also discusses publishing custom code and converted weights in separate Hub repositories, documenting usage and licenses, and optionally creating a Space for interactive testing. The full project took about 55 minutes for integration, with additional work for publishing and a Space, and was estimated to cost $33.07 in agent and cloud-compute charges. The approach is intended for personal use or sharing rather than an official Diffusers contribution, and loading custom code with `trust_remote_code=True` requires trusting the repository.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Opus 5.5 | 6 | 1,032 | 79 | 41 | +65% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.