Diffusers welcomes Stable Diffusion 3.5 Large — new model and quantization support

AI Impact Summary

Diffusers now supports Stable Diffusion 3.5 Large, offering improved image generation capabilities compared to Stable Diffusion 3. This release introduces key architectural changes like QK normalization and dual attention layers, optimized for large transformer models. Users can leverage this model through the Diffusers library, utilizing features like bfloat16 precision and quantization with bitsandbytes for memory optimization, particularly crucial for inference on consumer hardware.