Paper 2605.15178
SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformer
- Published
- May 2026
- Research lab
- NVIDIA
- Citations
- 13
- GitHub
- Not linked
01 In brief
Summary
SANA-WM is a 2.6B-parameter open-source world model for generating one-minute, 720p videos with precise 6-DoF camera control.
It uses a hybrid linear diffusion transformer combining frame-wise Gated DeltaNet (GDN) and softmax attention for efficient long-context modeling, a dual-branch camera control (UCPE and Plücker mixing), a two-stage generation pipeline with a long-video refiner, and a robust annotation pipeline that extracts metric-scale camera poses from public videos.
Trained on ~213K video clips in 15 days on 64 H100 GPUs, it generates 60-second clips on a single GPU, with a distilled variant running on an RTX 5090 in 34 seconds.
On a one-minute benchmark, SANA-WM achieves stronger action-following accuracy than prior open-source baselines and comparable visual quality at 36× higher throughput.
Key results include rotation error of 4.50° (Simple) and 8.34° (Hard) with the refiner, VBench Overall scores of 80.62/81.89, and revisit PSNR of 14.46/14.80 dB.
The architecture interleaves 15 GDN blocks with softmax attention at layers {3,7,11,15,19}, uses LTX2 VAE with 128 latent channels, and supports bidirectional, chunk-causal, and distilled autoregressive inference variants.
02 From the paper
Abstract
We introduce SANA-WM, an efficient 2.6B-parameter open-source world model natively trained for one-minute generation, synthesizing high-fidelity, 720p, minute-scale videos with precise camera control. SANA-WM achieves visual quality comparable to large-scale industrial baselines such as LingBot-World and HY-WorldPlay, while significantly improving efficiency. Four core designs drive our architecture: (1) Hybrid Linear Attention combines frame-wise Gated DeltaNet (GDN) with softmax attention for memory-efficient long-context modeling. (2) Dual-Branch Camera Control ensures precise 6-DoF trajectory adherence. (3) Two-Stage Generation Pipeline applies a long-video refiner to stage-1 outputs, improving quality and consistency across sequences. (4) Robust Annotation Pipeline extracts accurate metric-scale 6-DoF camera poses from public videos to yield high-quality, spatiotemporally consistent action labels. Driven by these designs, SANA-WMdemonstrates remarkable efficiency across data, training compute, and inference hardware: it uses only $\sim$213K public video clips with metric-scale pose supervision, completes training in 15 days on 64 H100s, and generates each 60s clip on a single GPU; its distilled variant can be deployed on a single RTX 5090 with NVFP4 quantization to denoise a 60s 720p clip in 34s. On our one-minute world-model benchmark, SANA-WM demonstrates stronger action-following accuracy than prior open-source baselines and achieves comparable visual quality at $36\times$ higher throughput for scalable world modeling.