01
Choose a still image
A clear face or character works best. Photos, illustrations, anime, and pets are all valid starting frames.
Free public demo
Wan 2.2 Speech to Video
Upload a photo and a voice track. Wan 2.2 S2V turns them into a talking, singing, or performance video — lip motion, expression, and body language included.
Visitor-free demo No API key 480P / 720P
Live tool
Generate a talking video
Image + audio in, MP4 out. The studio talks to the official Wan-AI Hugging Face Space so you can try S2V without a DashScope key.
Loading the official Space. Sleeping demos can take a minute to wake.
What it is
Wan 2.2 S2V is speech-to-video, not just lip sync
Wan 2.2 Speech to Video takes a still image and an audio clip, then generates a video where the subject talks, sings, or performs. It is the S2V variant in the Wan 2.2 family, aimed at talking photos, singing characters, and audio-driven avatars.
Unlike a mouth-only lip-sync tool, S2V uses the audio to drive a fuller performance: lips, expression, head motion, and often upper-body movement. Searchers looking for talking photo, singing photo, audio to video, AI lip sync, or talking avatar are usually describing this job.
How Wan 2.2 S2V works
Three inputs are enough for this demo: a picture, a soundtrack, and a resolution.
02
Add speech or singing
Video length follows the audio. Clean vocals give more readable lip motion than noisy background tracks.
03
Pick 480P or 720P
480P is faster for trying the feature. 720P costs more time on the public queue but looks sharper.
Image requirements
- Keep the main subject large and unobstructed.
- Front or three-quarter views beat extreme side profiles.
- One lead character is more reliable than a crowded group.
- Photos, artwork, and stylized characters are all acceptable.
Audio requirements
- Speech, singing, and performance clips are all in scope.
- Clear voice or vocal is more important than fancy mixing.
- Shorter clips are kinder to the free public queue.
- MP3, WAV, and M4A are the formats this demo expects.
What people actually generate
The same Image + Audio pipeline covers talking photos, singing performances, anime covers, and talking pets.
Talking photo
Give a portrait a spoken line. Useful for explainers, greetings, and talking-head avatars.
Singing photo
Pair a face with a vocal. The model treats singing as a performance, not a still mouth overlay.
Anime singing
Illustrated or anime characters can lip-sync and move with a song when the drawing is clear.
Pet talking
A pet photo plus a short voice clip is a common playful use — keep the animal large in frame.
S2V vs lip sync
If you only need the mouth to follow a script on an existing video, a lip-sync editor may be enough. S2V creates the video from a still image and audio.
| Feature | Wan 2.2 S2V | Typical lip sync |
|---|---|---|
| Motion | Audio-driven face, expression, and body performance | Mostly mouth shapes on an existing clip |
| Inputs | Still image + audio | Existing video + audio |
| Styles | Speech, singing, and performance | Usually speech only |
| Best for | Talking photos, singing characters, audio to video | Replacing dialogue on footage you already have |
S2V prompts and pairing tips
This public demo does not take a text prompt. The audio and image do the directing. Official APIs may add a style or prompt later; these pairings still help.
- 01. Match the energy: a calm voice with a calm portrait, a belted vocal with a performer pose.
- 02. Leave headroom around the face so generated head motion does not crop the subject.
- 03. Avoid busy backgrounds if you want the performance to stay readable.
- 04. One speaker in the audio is cleaner than overlapping voices.
Wan 2.2 S2V FAQ
Is Wan 2.2 S2V free here?
This page uses the official Wan-AI public Hugging Face Space, so visitors can try Image + Audio generation without an API key. It is a free demo backend, not a guaranteed production quota.
Does Alibaba DashScope still have a free S2V allowance?
No. Current DashScope pricing for Beijing-region wan2.2-s2v lists no free quota. Do not plan around the older “100 seconds free” note.
Do I need a GPU or my own key?
Not for this demo. The Space operator supplies the backend. Self-hosting the 14B Apache 2.0 weights is a different path and officially wants a very large GPU.
Why is my job queued or failing?
Public Spaces can sleep, rate-limit, or pause. Try 480P, shorter audio, the Official Space tab, or the ModelScope studio backup.
Is this the same as AI lip sync?
Related, but broader. Lip sync usually edits a mouth on existing video. S2V generates a new talking or singing video from a still image and audio.
Can I use this as a paid product backend?
Not as a stable production API. For real traffic, use a paid provider after you confirm demand. This page is for trying the feature and learning the workflow.