Efficient-Large-Model/SANA-WM_bidirectional · Hugging Face

Source: https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional

https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional#sana-wm-bidirectionalSANA-WM (Bidirectional)

SANA-WMis an efficient open-source world model trained natively for one-minute generation. The bidirectional checkpoint released here is a 2.6B-parameter image-to-video diffusion transformer that synthesises 720p, minute-scale videos with precise 6-DoF camera control, paired with the LTX-2 sink-bidirectional Euler refiner for high-fidelity decoding.

Four core designs drive the architecture:

1. Hybrid Linear Attention— frame-wise Gated DeltaNet combined with softmax attention every Nth block for memory-efficient long-context modelling.
2. Dual-Branch Camera Control— independent main and camera branches enable precise per-frame trajectory adherence.
3. Two-Stage Generation Pipeline— a long-video refiner stitched on top of Stage-1 latents improves quality and temporal consistency.
4. Robust Annotation Pipeline— metric-scale 6-DoF camera poses extracted from public video corpora yield spatiotemporally consistent action supervision.

Paper:https://arxiv.org/abs/2605.15178

@article{zhu2026sanawm,
  title   = {{SANA-WM}: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformer},
  author  = {Zhu, Haoyi and Liu, Haozhe and Zhao, Yuyang and Ye, Tian and Chen, Junsong and Yu, Jincheng and He, Tong and Han, Song and Xie, Enze},
  journal = {arXiv preprint arXiv:2605.15178},
  year    = {2026},
}

https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional#repository-layoutRepository layout

ComponentPath in repoSizeSana DiT (Stage 1)dit/sana\_wm\_1600m\_720p\.safetensors10 GBLTX-2 VAE (diffusers)vae/2 GBLTX-2 refiner (Stage 2)refiner/refiner\.safetensors41 GBGemma text encoder for the refinerrefiner/text\_encoder/46 GBInference configconfig\.yaml— The Sana text encoder (gemma\-2\-2b\-it) isnotbundled here — it is fetched on demand from the public Hugging Face mirror.

https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional#usageUsage

python inference_video_scripts/inference_sana_wm.py \
  --image      asset/sana_wm/demo_0.png \
  --prompt     asset/sana_wm/demo_0.txt \
  --action     "w-80,jw-40,w-40,lw-60,w-100" \
  --translation_speed 0.055 \
  --rotation_speed_deg 1.2 \
  --num_frames 321 \
  --output_dir results/demo

Weights are fetched from this repository on first use. Pass\-\-no\_refinerto skip the LTX-2 refiner and decode Stage-1 latents with the Sana VAE instead. To run fully offline, override any of\-\-config/\-\-model\_path/\-\-refiner\_checkpoint/\-\-refiner\_gemma\_rootwith local paths.

https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional#inputsInputs

ArgumentFormat\-\-imageRGB image (any PIL-readable format) — used as the first frame.\-\-promptUTF-8 text file containing the conditioning prompt.\-\-cameraNumPy\.npyof shape\(F, 4, 4\)— per-frame camera-to-world matrices.\-\-actionWASD/IJKL DSL, e.g."w\-80,jw\-40,w\-40,lw\-60,w\-100". We roll it out to a\(F\+1, 4, 4\)trajectory. Mutually exclusive with\-\-camera.\-\-intrinsicsOptional.\.npyof shape\(3, 3\),\(F, 3, 3\), or\(4,\). If omitted, we estimate intrinsics from\-\-imagewith Pi3X and abort if the resulting FOV is outside\[25°, 120°\]. The output frame size is fixed at704 x 1280; input images are aspect-preserving resized + center-cropped to that resolution.

https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional#licenseLicense

Released under the Apache 2.0 license. The bundled LTX-2 refiner and VAE inherit the LTX-2 upstream license.