v1.5.0
Henry-23/VideoChatv1.5.0Mar 25, 2025by lawrence-cj
AI Summary
A major update merging SANA-1.5 and SANA-Sprint, featuring significant performance improvements, support for 4K images, and new training capabilities.
Key Highlights
- Merged SANA-1.5 and SANA-Sprint releases
- Added support for 4K images with VAE-tiling to prevent Out Of Memory errors
- Integrated `diffusers` support for Sana models
- Added ControlNet and Sana-LoRA training capabilities
- Implemented FSDP training code and lower VRAM inference requirements
New Features
- Lower VRAM requirements for inference
- Fix config loading for Jupyter environments
- Fixed Dockerfile
- Fixed relative imports and triton dependencies
- Added ComfyUI related files and T2I2V video link
- Added diffusers support
- Added support for various image extensions in SanaImgDataset
- Added Sana-LoRA training and guidance
- Added config files for 2K models
- Updated code for better quality
- Fixed sample num bug in inference
- Fixed RNG state resume bug with different GPU numbers
- Added ImgDataset2WebDatasetMS tool
- Fixed 4K OOM with VAE-tiling
- Updated README
- Fixed CSS properties and documentation
- Fixed AutoencoderDC encode training bug
- Added 8-bit CAME optimizer code
- Added ControlNet support
- Added FSDP training code
- Added inference scaling update
- Added guidance exploration
- Added diffusers conversion script
Full Release Notes
## What's Changed * Inference with lower VRAM requirements by @frutiemax92 in https://github.com/NVlabs/Sana/pull/18 * Fix config loading for jupyter environments by @odusseys in https://github.com/NVlabs/Sana/pull/30 * Fix Dockerfile by @JimothyJohn in https://github.com/NVlabs/Sana/pull/34 * Relative import xformer triton depend fix by @lawrence-cj in https://github.com/NVlabs/Sana/pull/37 * remove triton dependency of DC-AE and fix bugs; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/38 * Refine gradio by @yujincheng08 in https://github.com/NVlabs/Sana/pull/39 * fix the bug of stop training around 4 hours. by @lawrence-cj in https://github.com/NVlabs/Sana/pull/66 * add comfyui related files by @lawrence-cj in https://github.com/NVlabs/Sana/pull/77 * add a T2I2V video link into comfyui readme by @lawrence-cj in https://github.com/NVlabs/Sana/pull/78 * 1. `diffusers` support Sana; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/88 * receive any kinds of image extension in `SanaImgDataset`; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/94 * add Sana-LoRA training and guidance by @lawrence-cj in https://github.com/NVlabs/Sana/pull/98 * 1. update for 2Kpx models release; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/103 * add config file for 2K models; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/104 * update code for better quality by @lawrence-cj in https://github.com/NVlabs/Sana/pull/105 * fix the sample num bug in inference. by @lawrence-cj in https://github.com/NVlabs/Sana/pull/108 * Nv labs GitHub repo/nv labs GitHub repo main by @lawrence-cj in https://github.com/NVlabs/Sana/pull/124 * fix the bug when resume rng_state with diffuerent gpu numbers. by @lawrence-cj in https://github.com/NVlabs/Sana/pull/131 * support 4K images by @lawrence-cj in https://github.com/NVlabs/Sana/pull/135 * [Tool]ImgDataset2WebDatasetMS by @Pevernow in https://github.com/NVlabs/Sana/pull/130 * fix 4K OOM with VAE-tiling by @lawrence-cj in https://github.com/NVlabs/Sana/pull/144 * docs: update README.md by @eltociear in https://github.com/NVlabs/Sana/pull/151 * fix: correct CSS property & missing articles by @CharlesCNorton in https://github.com/NVlabs/Sana/pull/152 * fix the bug for `AutoencoderDC` encode training; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/158 * Nv labs GitHub repo/nv labs GitHub repo main by @lawrence-cj in https://github.com/NVlabs/Sana/pull/163 * update CI script; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/165 * Sana1.5 8 bit CAME optimizer code release by @lawrence-cj in https://github.com/NVlabs/Sana/pull/167 * Nv labs GitHub repo/nv labs GitHub repo main adding controlnet (#23) by @lawrence-cj in https://github.com/NVlabs/Sana/pull/177 * update README by @lawrence-cj in https://github.com/NVlabs/Sana/pull/195 * SANA-1.5 && FSDP training code by @lawrence-cj in https://github.com/NVlabs/Sana/pull/196 * inference scaling update by @hills-code in https://github.com/NVlabs/Sana/pull/208 * Merging SANA-Sprint release code by @lawrence-cj in https://github.com/NVlabs/Sana/pull/198 * add a guidance to explore; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/212 * update diffusers conversion script; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/213 ## New Contributors * @frutiemax92 made their first contribution in https://github.com/NVlabs/Sana/pull/18 * @odusseys made their first contribution in https://github.com/NVlabs/Sana/pull/30 * @JimothyJohn made their first contribution in https://github.com/NVlabs/Sana/pull/34 * @lawrence-cj made their first contribution in https://github.com/NVlabs/Sana/pull/37 * @yujincheng08 made their first contribution in https://github.com/NVlabs/Sana/pull/39 * @Pevernow made their first contribution in https://github.com/NVlabs/Sana/pull/130 * @eltociear made their first contribution in https://github.com/NVlabs/Sana/pull/151 * @CharlesCNorton made their first contribution in https://github.com/NVlabs/Sana/pull/152 * @hills-code made their first contribution in https://github.com/NVlabs/Sana/pull/208 **Full Changelog**: https://github.com/NVlabs/Sana/commits/v1.5.0 ## What's Changed * Inference with lower VRAM requirements by @frutiemax92 in https://github.com/NVlabs/Sana/pull/18 * Fix config loading for jupyter environments by @odusseys in https://github.com/NVlabs/Sana/pull/30 * Fix Dockerfile by @JimothyJohn in https://github.com/NVlabs/Sana/pull/34 * Relative import xformer triton depend fix by @lawrence-cj in https://github.com/NVlabs/Sana/pull/37 * remove triton dependency of DC-AE and fix bugs; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/38 * Refine gradio by @yujincheng08 in https://github.com/NVlabs/Sana/pull/39 * fix the bug of stop training around 4 hours. by @lawrence-cj in https://github.com/NVlabs/Sana/pull/66 * add comfyui related files by @lawrence-cj in https://github.com/NVlabs/Sana/pull/77 * add a T2I2V video link into comfyui readme by @lawrence-cj in https://github.com/NVlabs/Sana/pull/78 * 1. `diffusers` support Sana; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/88 * receive any kinds of image extension in `SanaImgDataset`; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/94 * add Sana-LoRA training and guidance by @lawrence-cj in https://github.com/NVlabs/Sana/pull/98 * 1. update for 2Kpx models release; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/103 * add config file for 2K models; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/104 * update code for better quality by @lawrence-cj in https://github.com/NVlabs/Sana/pull/105 * fix the sample num bug in inference. by @lawrence-cj in https://github.com/NVlabs/Sana/pull/108 * Nv labs GitHub repo/nv labs GitHub repo main by @lawrence-cj in https://github.com/NVlabs/Sana/pull/124 * fix the bug when resume rng_state with diffuerent gpu numbers. by @lawrence-cj in https://github.com/NVlabs/Sana/pull/131 * support 4K images by @lawrence-cj in https://github.com/NVlabs/Sana/pull/135 * [Tool]ImgDataset2WebDatasetMS by @Pevernow in https://github.com/NVlabs/Sana/pull/130 * fix 4K OOM with VAE-tiling by @lawrence-cj in https://github.com/NVlabs/Sana/pull/144 * docs: update README.md by @eltociear in https://github.com/NVlabs/Sana/pull/151 * fix: correct CSS property & missing articles by @CharlesCNorton in https://github.com/NVlabs/Sana/pull/152 * fix the bug for `AutoencoderDC` encode training; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/158 * Nv labs GitHub repo/nv labs GitHub repo main by @lawrence-cj in https://github.com/NVlabs/Sana/pull/163 * update CI script; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/165 * Sana1.5 8 bit CAME optimizer code release by @lawrence-cj in https://github.com/NVlabs/Sana/pull/167 * Nv labs GitHub repo/nv labs GitHub repo main adding controlnet (#23) by @lawrence-cj in https://github.com/NVlabs/Sana/pull/177 * update README by @lawrence-cj in https://github.com/NVlabs/Sana/pull/195 ## New Contributors * @frutiemax92 made their first contribution in https://github.com/NVlabs/Sana/pull/18 * @odusseys made their first contribution in https://github.com/NVlabs/Sana/pull/30 * @JimothyJohn made their first contribution in https://github.com/NVlabs/Sana/pull/34 * @yujincheng08 made their first contribution in https://github.com/NVlabs/Sana/pull/39 * @Pevernow made their first contribution in https://github.com/NVlabs/Sana/pull/130 * @eltociear made their first contribution in https://github.com/NVlabs/Sana/pull/151 * @CharlesCNorton made their first contribution in https://github.com/NVlabs/Sana/pull/152 **Full Changelog**: https://github.com/NVlabs/Sana/commits/v1.0.0 ## What's Changed * SANA-1.5 && FSDP training code by @lawrence-cj in https://github.com/NVlabs/Sana/pull/196 * inference scaling update by @hills-code in https://github.com/NVlabs/Sana/pull/208 * Merging SANA-Sprint release code by @lawrence-cj in https://github.com/NVlabs/Sana/pull/198 * add a guidance to explore; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/212 * update diffusers conversion script; by @lawrence-cj in https://github.com/NVlabs/Sana/pull/213 ## New Contributors * @hills-code made their first contribution in https://github.com/NVlabs/Sana/pull/208 **Full Changelog**: https://github.com/NVlabs/Sana/compare/v1.0.0...v1.5.0