Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,9 @@ Join our [Discord](https://discord.gg/rde6eaE5Ta) to engage in discussions with

## News

- 🔥 [2026/10] 🎮 **StreamRender-H3** is released! A code-driven streaming render runtime connecting a controllable Three.js game engine to a multi-GPU H3 streaming renderer with browser video feedback. See [Project](https://nvlabs.github.io/Sana/Sol-Engine/StreamRender-H3/) | [Code](https://github.com/NVlabs/Sana/tree/sol-engine/models/streamrender_h3).
- 🔥 [2026/10] ⚡ **Sol-H3** brings MiniMax-H3 to a single RTX 5090 or DGX Spark, now also in ComfyUI: one 1344×768 clip with audio in **35.5 s** on one RTX 5090 (989.0 s for the original 50-step model). See [Project](https://nvlabs.github.io/Sana/Sol-H3/) | [Code](https://github.com/NVlabs/Sana/tree/sol-engine/models/minimax_h3).
- 🔥 [2026/09] 🔍 **SoL-Refiner** is released! One-step refinement turns low-resolution video drafts into high-resolution video, with code and models for LTX-2.3 and MiniMax-H3. See [Project](https://nvlabs.github.io/Sana/Sol-Refiner/) | [Code](https://github.com/NVlabs/Sana/tree/sol-engine/models/sol-refiner).
- 🔥 [2026/09] ⚡ **SANA-Video 2.0 5B 4-Step Preview** is released! The DMD preview generates 720p videos in four denoising steps and supports 5-second and 8-second outputs. See [Online Demo](https://huggingface.co/spaces/Efficient-Large-Model/sana-video2-5b-720p-demo) | [4-Step Weights](https://huggingface.co/Efficient-Large-Model/SANA-Video_2.0_5B_720p_4step) | [Doc](https://nvlabs.github.io/Sana/docs/sana_video2/) | [Project](https://nvlabs.github.io/Sana/Video2/).
- 🔥 [2026/08] 🎬 **SANA-Video 2.0** training, inference, model architecture, and 5B 720p checkpoint are released! The 8-second model supports both text-to-video and text-image-to-video generation, with hybrid linear/softmax attention and Attention Residuals. See [Online Demo](https://huggingface.co/spaces/Efficient-Large-Model/sana-video2-5b-720p-demo) | [Project](https://nvlabs.github.io/Sana/Video2/) | [Doc](https://nvlabs.github.io/Sana/docs/sana_video2/) | [Model Zoo](https://nvlabs.github.io/Sana/docs/model_zoo/#sana-video-20) | [Weights](https://huggingface.co/Efficient-Large-Model/SANA-Video_2.0_5B_720p).
- 🔥 [2026/08] ⚡ **Sol Engine: Day-One MiniMax-H3 Acceleration** is available! The 33B omni-modal audio+video DiT runs **3.95×** faster on GB200, reached in 4.5 hours of optimization, and up to **4.52×** on hardware that sits on a desk — 3.92× on DGX Spark, 4.52× on GeForce RTX 5090 — with no distillation, no LoRA, and no calibration pass. See [GB200 Blog](https://nvlabs.github.io/Sana/Sol-Engine/H3/) | [On-Device Blog](https://nvlabs.github.io/Sana/Sol-Engine/H3-OnDevice/).
Expand Down
4 changes: 4 additions & 0 deletions docs/index.md
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,10 @@ ______________________________________________________________________
| **Sana-Video** | Efficient video generation with Block Linear Attention |
| **Sana-Video 2.0** | 5B/14B hybrid-attention video models with Attention Residuals; [project page](https://nvlabs.github.io/Sana/Video2/), [4-step preview](https://huggingface.co/spaces/Efficient-Large-Model/sana-video2-5b-720p-demo), and [50-step](https://huggingface.co/Efficient-Large-Model/SANA-Video_2.0_5B_720p) / [4-step](https://huggingface.co/Efficient-Large-Model/SANA-Video_2.0_5B_720p_4step) weights released |
| **LongSana** | Minute-length real-time video generation (with LongLive) |
| **SANA-WM** | 2.6B world model for 720p, minute-scale video with 6-DoF camera control; [docs](sana_wm.md) |
| **SANA-Streaming** | 2B real-time streaming video editing at 720p; [docs](sana_streaming.md) |
| **Sol-RL** | NVFP4-rollout, BF16-training diffusion RL; [docs](sol_rl.md) |
| **Sol-Engine** | Inference acceleration for video diffusion models; [docs](https://nvlabs.github.io/Sana/Sol-Engine/docs/) |

### Key Techniques

Expand Down
20 changes: 19 additions & 1 deletion docs/model_zoo.md
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@
| LongSANA-Video-2B-ODE-Init | 480p | [LongSANA_2B_480p_ode](https://huggingface.co/Efficient-Large-Model/LongSANA_2B_480p_ode) | --- | bf16 | LongSANA first step model initialized from ODE trajectories |
| LongSANA-Video-2B-Self-Forcing | 480p | [LongSANA_2B_480p_self_forcing](https://huggingface.co/Efficient-Large-Model/LongSANA_2B_480p_self_forcing) | --- | bf16 | LongSANA second step model trained by Self-Forcing |

### [SANA-Video 2.0](https://nvlabs.github.io/Sana/Video2/)
### [SANA-Video 2.0](https://arxiv.org/abs/2607.21553)

[Documentation](sana_video2.md) · [4-step preview](https://huggingface.co/spaces/Efficient-Large-Model/sana-video2-5b-720p-demo)

Expand All @@ -52,6 +52,24 @@
| SANA-Video 2.0 5B 4-step preview | 720p | [SANA-Video_2.0_5B_720p_4step](https://huggingface.co/Efficient-Large-Model/SANA-Video_2.0_5B_720p_4step) | --- | bf16 inference | T2V-only DMD preview (81 frames at 16 FPS or 193 frames at 24 FPS) |
| SANA-Video 2.0 14B | 480p | Coming soon | Coming soon | bf16 | 40-layer hybrid-attention model |

### [SANA-WM](https://arxiv.org/abs/2605.15178)

[Documentation](sana_wm.md)

| Model | Reso | pth link | Description |
| --- | --- | --- | --- |
| SANA-WM 2.6B bidirectional | 720p | [SANA-WM_bidirectional](https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional) | 1-min world model with 6-DoF camera control |
| SANA-WM streaming | 720p | [SANA-WM_streaming](https://huggingface.co/Efficient-Large-Model/SANA-WM_streaming) | Streaming (chunk-causal) variant |

### [SANA-Streaming](https://arxiv.org/abs/2605.30409)

[Documentation](sana_streaming.md)

| Model | Reso | pth link | Description |
| --- | --- | --- | --- |
| SANA-Streaming 2B | 720p | [SANA-Streaming](https://huggingface.co/Efficient-Large-Model/SANA-Streaming) | 1-min streaming video editing |
| SANA-Streaming 2B bidirectional | 720p | [SANA-Streaming_bidirectional](https://huggingface.co/Efficient-Large-Model/SANA-Streaming_bidirectional) | 5-second bidirectional video editing |

______________________________________________________________________

## ❗ 2. Make sure to use correct precision(fp16/bf16/fp32) for training and inference.
Expand Down
16 changes: 15 additions & 1 deletion docs/sana_video2.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@ and 720p reference configs for the 5B variant, and a post-trained 5B checkpoint
for 720p, 8-second generation. The 14B config and checkpoint are not included
yet.

> **Release status:** Explore the [project page](https://nvlabs.github.io/Sana/Video2/),
> **Release status:** Read the [paper](https://arxiv.org/abs/2607.21553), explore the [project page](https://nvlabs.github.io/Sana/Video2/),
> try the [4-step 5B 720p preview](https://huggingface.co/spaces/Efficient-Large-Model/sana-video2-5b-720p-demo),
> or download the [50-step](https://huggingface.co/Efficient-Large-Model/SANA-Video_2.0_5B_720p)
> and [4-step preview](https://huggingface.co/Efficient-Large-Model/SANA-Video_2.0_5B_720p_4step)
Expand Down Expand Up @@ -237,3 +237,17 @@ torchrun --nproc_per_node=8 --master_port=29500 \
To resume, pass `--resume_from=<checkpoint>`. Set `model.load_from` when
initializing from a compatible model checkpoint; leave it as `null` to train
from scratch.

## Citation

```bibtex
@misc{chen2026sanavideo2,
title={SANA-Video 2.0: Hybrid Linear Attention with Attention Residuals for Efficient Video Generation},
author={Junsong Chen and Jincheng Yu and Yitong Li and Shuchen Xue and Haozhe Liu and Jingyu Xin and Yuyang Zhao and Tian Ye and Zhangjie Wu and Zian Wang and Daquan Zhou and Ping Luo and Song Han and Enze Xie},
year={2026},
eprint={2607.21553},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2607.21553},
}
```
2 changes: 1 addition & 1 deletion mkdocs.yml
Original file line number Diff line number Diff line change
Expand Up @@ -104,4 +104,4 @@ extra_css:
- stylesheets/extra.css
extra_javascript:
- stylesheets/extra.js
copyright: Copyright &copy; 2024-2025 Sana Team
copyright: Copyright &copy; 2024-2026 Sana Team
Loading