Diffusers 0.35.0 adds Wan 2.2, Flux-Kontext, Qwen-Image pipelines and faster loading

AI Impact Summary

Diffusers 0.35.0 introduces Wan 2.2, Flux-Kontext, Qwen-Image, and Qwen-Image-Edit pipelines, plus a new video pipeline and training scripts, signaling a substantial expansion of image/video generation capabilities. The release also includes an attention refactor enabling multiple backends, regional compilation to reduce cold-start times, and faster pipeline loading on accelerator devices, which collectively improve performance and deployment efficiency for large models like Wan and Qwen-Image. For teams, this means easier customization via single-file modeling, experimental Modular Diffusers, and GGUF-enabled speedups, but requires consulting docs and potential code changes to adopt new pipelines and training workflows.