WAN 2.2 Speech-to-Video
Turn a portrait and an audio clip into a talking video with WAN 2.2 Speech-to-Video. 480p or 720p, up to 60s, billed by audio length from 20 credits/s.
WAN 2.2 Speech-to-Video turns a still portrait and an audio track into a talking performance. It synchronises lip movement to the supplied voice and adds natural facial expression, body movement and camera framing. Upload WAV or MP3 up to 15MB and 60 seconds; output is 480p or 720p, billed by the audio's length at 20 or 40 credits per second.
- Input: Image + audio
- Audio: WAV / MP3 ≤15MB
- Price: 20–40 credits/s
- Max length: 60s
How to use WAN 2.2 Speech-to-Video
The audio is the script. Clean speech in gives clean lip sync out; the image supplies the face, not the performance.
- Pick a clean, front-facing portrait The model needs visible lips and eyes. Sharp focus and even lighting matter more than a dramatic pose.
- Upload the voice track Use WAV or MP3 up to 15 MB and 60 seconds. Trim silence at the start and end so the first frame matches the first word.
- Add optional guidance A short prompt can set the framing or expression: look at the camera, warm smile, keep the head still. Leave it empty for a neutral read.
- Choose the resolution 480p is enough to check sync; 720p is the delivery tier. Close-ups benefit most from the extra detail.
- Check the first and last frames Sync drift usually appears at the edges. If the mouth is ahead or behind at the start, trim the audio and rerun.
Settings reference
WAN 2.2 Speech-to-Video turns a portrait and an audio track into a talking performance, with lip sync driven by the supplied voice.
| Setting | Value | Notes |
|---|---|---|
| Portrait image | 1 image, required | A clear, front-facing photo with an unobstructed mouth and eyes. |
| Audio | Required, WAV/MP3 up to 15 MB | Up to 60 seconds. The audio length drives the output length and the bill. |
| Prompt | Optional | Additional guidance for framing, expression or delivery. |
| Resolution | 480p / 720p | 20 credits per second at 480p, 40 at 720p. 480p is the default. |
| Seed | Optional | Reuse a seed to reproduce a performance while changing the audio. |
| Billing | 3s – 60s | Billed by the audio's length, with a 3-second minimum and 60-second cap. |
What the model animates
Speech-to-Video is narrower than a general image animator. Its job is to make a face speak convincingly: lips, jaw, subtle head movement and expression follow the supplied audio. It also adds body movement and a professional camera frame, so the result reads as a shot rather than a talking-head cutout.
Because the performance is driven by audio timing, the voice is not generated or cloned here. You bring the track — a recorded voiceover, a TTS render or a licensed clip — and the model fits the face to it. That makes it useful for localisation: keep one portrait and swap the audio track for each language.
- Lip sync driven by uploaded audio
- Facial expression and head movement
- Body movement and camera framing
- Voice is supplied, not generated
- Portrait remains consistent
Pricing by audio length
Billing follows the audio duration, not the output video's nominal length. 480p costs 20 credits per second and 720p costs 40, with a 3-second minimum and a 60-second maximum. There is also a 60-credit minimum charge, which matters for very short clips.
A 5-second 480p clip costs 100 credits; a 30-second 720p clip costs 1,200. Because cost scales linearly with speech, trimming silence and tightening the script before generation is the simplest way to reduce spend.
- 20 credits per second at 480p
- 40 credits per second at 720p
- Minimum billed length: 3 seconds
- Maximum: 60 seconds
- 60-credit minimum charge
| Audio length | 480p | 720p |
|---|---|---|
| 3s | 60 | 120 |
| 5s | 100 | 200 |
| 10s | 200 | 400 |
| 30s | 600 | 1,200 |
| 60s | 1,200 | 2,400 |
Preparing the portrait
Use a front-facing or slightly angled portrait with a clear view of the mouth and eyes. Even lighting across the face matters more here than in general image-to-video, because shadows can be read as lip or jaw artefacts. Avoid heavy occlusion: hands, microphones, scarves and hair across the mouth all reduce sync quality.
A neutral expression gives the model room to animate. A wide smile or a mid-speech expression can lock the face into a shape that fights the new audio. Match the portrait's framing to the intended shot — head-and-shoulders for a presenter, waist-up for a gesturing character — and leave some headroom.
- Front-facing or slight angle
- Mouth and eyes clearly visible
- Even, soft lighting
- Neutral starting expression
- Head-and-shoulders or waist-up
Preparing the audio
Use clean speech with minimal background noise. The model synchronises to what it hears, so music beds and room echo can blur the timing. If the track is a mix, export the voice stem first. Keep the file under 15MB and within 60 seconds; WAV avoids a second lossy encode, while MP3 is fine for a draft.
Normalise the level before upload and leave short pauses at the start and end. A track that begins mid-word gives the model no room to settle the face. For multilingual versions, generate each language as a separate run rather than concatenating tracks.
- Clean voice, minimal background noise
- Export the voice stem from a mix
- WAV or MP3, ≤15MB
- Normalise level, keep lead-in silence
- One language per run
Prompts, limits and alternatives
The prompt is optional. Use it for framing or mood guidance — presenter at a desk, warm office light, calm delivery — rather than for scripting the speech, which is already in the audio. A short prompt is less likely to pull the face away from the source portrait.
The endpoint stops at 720p and 60 seconds, produces a silent video track by design (the audio is the input, not an output), and offers no last-frame control. For a cinematic still animation without speech, use HappyHorse Image-to-Video. For motion copied from a performance video, use WAN 2.2 Animate.
- Prompt optional
- 720p maximum
- 60-second maximum
- No generated voice
- Animate for motion, HappyHorse for cinematic stills
Prompt examples for WAN 2.2 Speech-to-Video
- Presenter at a desk: A presenter speaks directly to camera at a wooden desk, warm office light, medium shot, calm delivery, subtle hand gestures. Keep the framing stable.
- Product explainer: A woman in a bright studio speaks to camera beside a product on a plinth, soft key light, medium close-up, friendly expression. Avoid covering the mouth with the product.
- Character dialogue: A fantasy character in a stone hall speaks with a serious expression, torchlight from the left, shoulder-up framing, slight head movement. Match the portrait's existing lighting.
- Localised ad: A young man speaks to camera in a modern kitchen, morning light, chest-up framing, natural gestures, clean commercial look. Run once per language track.
- Teacher explainer: A teacher speaks to camera in front of a whiteboard, even classroom light, waist-up framing, relaxed posture. Leave room for hand movement.
- Podcast clip: A host speaks into a studio microphone, dark acoustic panels behind, warm side light, close-up framing, engaged expression. Keep the microphone out of the mouth area.
Pro tips for WAN 2.2 Speech-to-Video
This is a talking-head tool. Keep the frame simple and let the audio carry the performance.
- Clean audio beats long audio Background noise and overlapping voices confuse the sync. Run a noise reduction pass before uploading if the recording is rough.
- The 15 MB and 60-second caps Both limits bind quickly with uncompressed WAV. Export MP3 at a sensible bitrate so a full minute fits under 15 MB.
- Trim silence first Leading silence produces a still face for the first second. Tightening the track makes the output feel immediate.
- Keep the head in frame Large pose changes are not this model's job. A stable portrait with natural expression reads as a real performance.
- 720p for delivery, 480p for checks Draft the sync at 480p at half the price, then rerun the approved take at 720p.
Frequently asked questions
How much does WAN 2.2 Speech-to-Video cost?
20 credits per second at 480p and 40 credits per second at 720p, billed on the audio length. A 5-second 480p clip is 100 credits.
What audio formats are supported?
WAV and MP3, up to 15MB. The track can be up to 60 seconds long.
What is the minimum charge?
Billing starts at 3 seconds, and the model has a 60-credit minimum charge.
Does it clone or generate a voice?
No. You supply the audio; the model animates the face to match it. Use MiniMax Voice Clone or Speech models to create the voice separately.
What resolutions are available?
480p and 720p.
Is a prompt required?
No. An optional prompt can guide framing and mood, but the speech comes from the audio.
Why is the lip sync drifting?
Usually because the audio has background music, echo or an unclear voice. Export a clean voice stem, normalise it and leave a short pause at the start.
How does it differ from WAN 2.2 Animate?
Speech-to-Video drives a portrait from audio. Animate drives a character from a motion video. Use Speech-to-Video for talking performances and Animate for copied body movement.
Compare with other talking-video models
- WAN 2.2 Animate
- WAN 2.2 Image-to-Video
- Kling Image-to-Video
- HappyHorse 1.0 Image-to-Video
- MiniMax Speech 2.8 Turbo
- Browse all AI video models
Specifications
- Provider: Wavespeed
- Model type: Video to video
- Credit pricing: 480p: 20 credits · 720p: 40 credits
- Capabilities: popular, cinematic, realistic, audio-sync
About Conut.ai
Conut.ai is an AI image, video and audio generation platform with 50+ creative tools. It supports text-to-image, image-to-video, face swap, motion control, relighting, inpainting and video effects through models including Nano Banana Pro, Seedream, GPT Image, Sora 2, Kling, WAN and Seedance. Start free — no credit card required.