Hugging Face added Stable Diffusion 3.5 Large to its Diffusers library on April 5. The model generates images at 4K resolution. It uses a new open-source license.
4K Output Support
The model produces images up to 3840x2160 pixels. Previous versions maxed at 2048x1024. This doubles the pixel count. Artists and designers can now export for cinema-grade workflows. No GPU upgrades are required for existing users.
License Change
Stable Diffusion 3.5 Large uses the Apache 2.0 license. Prior versions had the CC-BY-NC 4.0 license. This allows commercial use without attribution. Researchers can now apply the model in closed-source products. The change aligns with Meta's Llama 3 licensing.
Performance Benchmarks
Inference latency averages 1.2 seconds per image. That's 18% faster than Stable Diffusion 3.0. Testing used A100 GPUs at 16-bit precision. The team optimized attention mechanisms and reduced layer count.
Hugging Face plans to release a 8K resolution variant by Q4 2024. Source: Hugging Face Blog