SANA-Video 2.0 · 5B · 720p · 4-Step Preview

Project page · 4-step model · 50-step base model · Documentation · GitHub · Paper

This Space serves the independent four-step distilled SANA-Video 2.0 5B 720p preview checkpoint. It generates 736×1280 videos with four sampling steps, using a five-second (81 frames at 16 FPS) or eight-second (193 frames at 24 FPS) temporal profile. The Space uses the latest global-step-1000 tri-temporal DMD EMA export and exposes duration, motion conditioning, and RL LoRA strength controls. The original 50-step base model remains available for the full T2V and TI2V pipelines and is not replaced by this T2V-only preview.

Exact preview settings

Each non-terminal stage predicts x0 and re-noises it at the next fixed sigma with a fresh draw from the same seeded generator. flow_shift is not applied by this sampler. Resolution, CFG, and the four-step schedule stay fixed; the UI exposes the validated duration profiles, Motion Score, RL LoRA scale, and seed.

Pipeline

The transformer is the latest global-step-1000 full-model EMA export distilled from the ReFL step-500 initialization. It is not replaced by a LoRA and contains no optimizer, scheduler, or training-loop state. The separately published rank-128 RL LoRA is used only to adjust RL strength around that full model at runtime:

W(s) = W(scale=1) + (s - 1) × (alpha / rank) × (B_RL @ A_RL)

Scale 1.0 reproduces the uploaded full checkpoint. The default 0.7 setting is a post-hoc adjustment around that reference, not a separately DMD-trained model. Every scale change is reconstructed from immutable BF16 reference weights, so repeated slider changes do not accumulate rounding error.

Vendored code

The diffusion/ package in this Space is NVlabs/Sana (Apache-2.0), pruned to the modules needed for SANA-Video 2.0 inference. FastVideoDMD4Step mirrors the stochastic four-stage rollout used in training. Unit tests compare its timesteps, RNG consumption, transitions, trace metadata, and BF16 behavior with a manual reference implementation.

The tiny mmcv/ package is a compatibility stand-in that provides only the four symbols touched by the vendored inference code (Registry, build_from_cfg, Config/ConfigDict, and get_dist_info).

Attribution

The five text-to-video examples are selected from the official SANA-Video 2.0 project page curated set, hosted in Efficient-Large-Model/Sana-assets.