All Tools
S
OtherFreeOpen Source
SANA
Efficient linear DiT for image and video generation
Apache-2.0
ABOUT
High-res diffusion is slow and VRAM-heavy, especially for video. SANA replaces quadratic attention with linear DiT blocks so 1024 and 4K images, few-step Sprint sampling, and 720p video fit on a single GPU with open training and serving code.
INTEGRATION GUIDE
1. Generate 1024 or 4K images with SANA Diffusers pipelines or ComfyUI
2. Run SANA-Sprint one-step sampling on a 4090 or H100
3. Train or sample SANA-Video 2.0 720p text-to-video and image-to-video
4. Serve SANA through SGLang with an OpenAI-compatible API
TAGS
pythondiffusiontext-to-imagevideonvidiadiffuserscomfyui