Speech-to-video

Official generate.py

python generate.py --task s2v-14B --size 1024*704 --ckpt_dir ./Wan2.2-S2V-14B/ --offload_model True --convert_model_dtype --prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard." --image examples/i2v_input.JPG --audio examples/talk.wav

python generate.py --task s2v-14B --size 1024*704 --ckpt_dir ./Wan2.2-S2V-14B/ --offload_model True --convert_model_dtype --image examples/i2v_input.JPG --enable_tts --tts_prompt_audio examples/zero_shot_prompt.wav --tts_prompt_text "…" --tts_text "…"

Single-GPU official floor is 80 GB. Extra extras: --pose_video for pose-driven singing/acting; --num_clip for a short preview. Install requirements_s2v.txt if you use CosyVoice. Size is an area budget; aspect follows the image.

Published comparisons (project page)

The Wan-S2V page reports FID 15.66, SSIM 0.734, PSNR 20.49, CSIM 0.677 — best among the listed talking-head baselines on those metrics — with competitive Sync-C / EFID. Treat the table as the authors’ measurement, not a universal leaderboard.

FAQ

Is S2V native audio-video?

No. You supply (or CosyVoice-synthesize) the audio. The model animates picture to that track. Native joint AV is Wan 2.5+ APIs.

Diffusers?

S2V is listed as integrated in the Wan2.2 todo (inference, ComfyUI, Diffusers). Prefer the official README flags if a wrapper lags.