Gallery
Inference time = how long it takes to make one 81-frame video, autoencoder included. Measured on one A100 80GB at 50 steps, CFG 5, batch 1, bf16.
Prompts shown are the VBench captions the videos are scored against.
No videos match this combination. Try clearing one of the filters.
Wan2.1-14B vs. GRACE
Thumbnails show GRACE. Hover to compare with the pretrained Wan2.1-14B.More comparisons
Wan2.1-14B, GRACE, and a third baseline: DC-Gen, or LTX-Video 0.9.7 on the I2V samples where we compare against it.More GRACE samples
Hover to enlarge.