Diffusers 0.41.0 adds QwenImage 2.1 and changes ONNX support

The library update adds image tools and introduces compatibility changes for developers upgrading older projects.

Marcus Reid
October 6, 2026
Tools
1 min

Laptop and desktop computer with dark screens on a tidy desk beside a window — Illustrative archival photograph by Norbert Levajsics under CC0

Listen to this article

Hugging Face released Diffusers 0.41.0 on October 6, adding QwenImage 2.1 support to the library. Developers can generate images from text, edit existing pictures, use multiple reference images and produce images with transparent backgrounds. The integration also supports LoRA training.

The update changes how tensor parallel loading handles model weights. Each process reads its own portion of a checkpoint, which the maintainers say reduces memory use during loading. Modular Diffusers also avoids downloading weights when loading from an existing local path.

There is an upgrade catch for teams using ONNX. Its support is deprecated, with the maintainers directing ONNX Runtime users to Optimum. Older Lumina pipeline aliases have also been removed, so existing integrations need checking before an upgrade.

Illustrative archival photograph by Norbert Levajsics under CC0. Resized and converted to WebP.

Marcus Reid

Marcus Reid is focused on covering the money, rules, and institutional choices shaping AI. He runs from funding rounds and chip deals to regulation, lawsuits, leadership changes, and the business of building enormous computing systems. Marcus follows the incentives behind the announcement. Who pays, who gains leverage, and what changes for everyone else? The voice is direct, measured, and occasionally dry, especially when a grand promise arrives with very little detail.