WAN 2.5 Image-to-Video
Animate a photo with WAN 2.5 image-to-video: one-pass audio and lip sync, 480p–1080p, 5–10s clips, 20–40 credits/s. Upload a voice track or let the model generate one.
WAN 2.5 Image-to-Video is the first WAN checkpoint on Conut.ai that generates picture and sound together. Upload a photo, write the scene and any spoken line, and the model returns a 5- or 10-second clip with synchronized audio and lip movement. It supports 480p, 720p and 1080p, understands prompts in multiple languages, and accepts an optional voice track when you need a specific voice. It costs 20 credits per second at 480p, 30 at 720p and 40 at 1080p.
- Resolution: 480p / 720p / 1080p
- Duration: 5s / 10s
- Price: 20–40 credits/s
- Audio: One-pass A/V sync
How to use WAN 2.5 Image-to-Video
Plan the line before the picture. One-pass audio/video sync means the dialogue length and the clip length have to agree.
- Choose a front-facing source image The model needs visible mouth and face detail to synchronize speech. Profile shots and heavy occlusion produce weak lip movement.
- Write the scene and the exact line Put the spoken words in the prompt, in quotes if you like, and keep them short enough for the chosen duration. Roughly two to three words per second is the safe ceiling.
- Add an audio track when the voice must be exact Upload a WAV or MP3 up to 15 MB if you already have the recording. Without an upload, the model generates the voice from the prompt.
- Pick the duration and resolution Five seconds covers a greeting or a product line; ten seconds fits a short explanation. 1080p costs 40 credits per second and is worth it for close-ups.
- Check lip sync in the first two seconds If the mouth is out of step at the start, it rarely recovers. Rerun with a tighter line or a clearer source image rather than adding prompt detail.
Settings reference
WAN 2.5 Image-to-Video adds synchronized sound to image animation. The prompt carries both the scene and the spoken line.
| Setting | Value | Notes |
|---|---|---|
| Input image | 1 image, required | A clear portrait or product shot; the model keeps identity while adding motion and speech. |
| Prompt | Required | Describe the scene and include the exact line you want spoken. |
| Audio upload | Optional, up to 15 MB | Supply a voice track when the delivery must match a specific recording. |
| Duration | 5s / 10s | Default is 5 seconds. Longer dialogue needs the 10-second option. |
| Resolution | 480p / 720p / 1080p | 20, 30 or 40 credits per second respectively. 720p is the default. |
| Negative prompt | Optional | Useful for warped faces and lip artefacts in close-ups. |
| Prompt expansion | Optional toggle | Expands sparse prompts; leave it off when you are supplying exact dialogue. |
| Seed | Optional | Reuse a seed to keep a performance while you adjust the script. |
What one-pass audio and video sync means
Earlier WAN image-to-video models returned silent clips. WAN 2.5 generates the voice, ambient sound and lip movement in the same pass as the picture, so the audio is driven by the scene rather than added afterwards. If you write dialogue in quotes, the model uses it as the spoken line and times the mouth movement to match.
The model can also work without a script. If you do not provide audio, it can generate a voice or leave the clip silent depending on the prompt. That makes it useful for talking-head photos, presenter shots, product voiceovers and short social clips where adding sound in an editor would otherwise be the slow part.
- Dialogue in quotes becomes the spoken line
- Lip movement is generated with the video, not post-synced
- Ambient sound can be described in the prompt
- Optional audio upload for a specific voice or music bed
WAN 2.5 vs WAN 2.2 vs WAN 2.6
WAN 2.5 sits between the cheaper 2.2 models and the longer, higher-resolution 2.6 model. It is the one to pick when the subject needs to speak or when the clip needs sound but does not need to run 15 seconds.
- WAN 2.5 is cheaper than WAN 2.6 at every shared resolution.
- WAN 2.5 keeps 480p, which is useful for cheap drafts of a talking shot.
- WAN 2.6 adds 15-second clips and a higher top resolution, but no 480p tier.
- Both WAN 2.5 and 2.6 generate sound; only 2.5 accepts an uploaded audio track here.
| Model | Resolution | Duration | Credits | Audio | Best for |
|---|---|---|---|---|---|
| WAN 2.2 I2V | 480p / 720p | 5s, 8s | 30–60 /s | No | Controlled silent motion |
| WAN 2.2 Spicy | 480p / 720p | 5s, 8s | 20–40 /s | No | Expressive silent motion |
| WAN 2.5 I2V | 480p / 720p / 1080p | 5s, 10s | 20–40 /s | Yes, one-pass A/V sync | Talking shots, voiceover, lip sync |
| WAN 2.6 I2V | 720p / 1080p | 5s, 10s, 15s | 30–50 /s | Yes | Longer cinematic clips |
Writing dialogue for lip sync
Put the spoken line in quotes and keep it short. A 5-second clip fits roughly one sentence; a 10-second clip fits two or three. Give the model a voice direction in the same prompt: calm narrator, excited host, whispered aside, formal newsreader. If the line is too long, the model either rushes it or the lip movement drifts.
For multilingual clips, write the prompt and the dialogue in the target language. WAN 2.5 is tuned for Chinese and English, and it handles other languages with varying accuracy. Short, common words sync better than long technical terms or names.
- Quote the exact line: She says: “Welcome back to the studio.”
- Add a voice direction: warm, low, energetic, whispered, formal.
- Keep one speaker per clip; multiple speakers are hard to time.
- Write non-English dialogue in the target language, not a translation in the prompt.
- Avoid long numbers, URLs and brand names when the mouth must match the words.
Using your own audio track
The optional audio upload accepts WAV or MP3 files from 3 to 30 seconds, up to 15 MB. If the audio is longer than the selected video duration, only the matching first segment is used. If it is shorter, the remaining video plays silently. Upload a clean voice track without music if the priority is lip sync; a busy mix gives the model less to lock onto.
- Formats: WAV or MP3, 3–30 seconds, 15 MB maximum
- Trim the track to the target duration before uploading
- Use a dry voice track for lip sync, add music in the editor afterwards
- If the voice starts late, add a short silence at the start rather than shortening the clip
Limitations and when to use another model
WAN 2.5 caps clips at 10 seconds and does not expose a 15-second duration. At 1080p it is the most expensive WAN 2.5 tier, so a long batch of drafts is cheaper at 480p. The model still struggles with fast hand gestures, profile views that hide the mouth, and scenes where the face is small in frame. It also cannot guarantee perfect lip sync when the source image has an unusual head angle or the mouth is partly covered.
- Need 15 seconds? Use WAN 2.6 image-to-video.
- Need a silent, cheaper draft? Use WAN 2.2 Spicy at 480p.
- Need a negative prompt? WAN 2.5 supports one, unlike Spicy.
- Keep the face large and the head angle close to the camera for the strongest lip sync.
- For product b-roll without speech, a silent model is usually cheaper and just as good.
Prompt examples for WAN 2.5 Image-to-Video
- Conference keynote: A confident woman in her 40s stands on a stage with a microphone, warm spotlight, dark auditorium behind her. She smiles and says: “Good evening everyone. Tonight I want to share three lessons about leadership.” Calm, confident voice, slow push-in, cinematic corporate lighting. One sentence for a 5-second clip; two or three for 10 seconds.
- Product voiceover: A glass skincare bottle on a wet marble counter, morning light, water droplets on the glass. A warm female voice says: “Hydration that lasts all day.” Slow orbit around the bottle, soft studio reflections, premium commercial style. The product stays still while the voice carries the message.
- Multilingual presenter: A young man in a white shirt stands in front of a blurred city street, direct eye contact with the camera. He says in Mandarin: “欢迎来到今天的产品演示。” Natural hand gesture, soft daylight, shallow depth of field, documentary style. Write the dialogue in the target language, not an English translation.
- UGC selfie: A woman filming herself in a car, daylight through the windshield, slight handheld camera movement. She laughs and says: “I did not expect this to work so well.” Casual, energetic voice, natural skin texture, social media framing. Handheld framing and a casual voice direction sell the UGC look.
- Historical portrait: A weathered sea captain in a wool coat stands on a wooden deck, fog behind him, lantern light on his face. He says slowly: “The storm is coming.” Low, gravelly voice, wind and creaking wood in the background, cinematic period drama. Describe ambient sound separately from the spoken line.
- Customer testimonial: A woman in a bright kitchen leans toward the camera, coffee cup in hand. She says: “I switched last month and I am not going back.” Friendly, conversational voice, soft morning window light, clean lifestyle commercial. Keep the line conversational; long marketing copy rushes at 5 seconds.
Pro tips for WAN 2.5 Image-to-Video
Audio sync is the reason to choose 2.5 over 2.2. Write for the ear, not just the eye.
- One speaker per shot Two people talking in one 5-second clip splits the model's attention and produces mushy lip movement. Cut between single-speaker shots instead.
- Keep dialogue under the duration A 10-second clip cannot fit 30 words comfortably. Write to the runtime, then trim a beat rather than letting the last words rush.
- Upload audio for brand voices Generated voices are convenient but not contract-safe. When a specific recording or voice actor is required, upload the track.
- Turn expansion off for exact scripts Prompt expansion can rewrite phrasing, which changes what the model tries to say. Disable it when the line is fixed copy.
- Use 1080p for faces Lip detail disappears at 480p on a close-up. Draft at 720p, then deliver the chosen take at 1080p.
Frequently asked questions
Does WAN 2.5 image-to-video generate audio?
Yes. It generates picture and sound in one pass, including voice and lip movement. You can also upload a WAV or MP3 track from 3 to 30 seconds (up to 15 MB) when you need a specific voice.
How much does WAN 2.5 image-to-video cost?
480p costs 20 credits per second, 720p costs 30 and 1080p costs 40. A 5-second 720p clip is 150 credits and a 10-second 1080p clip is 400 credits.
How long can a WAN 2.5 clip be?
The endpoint supports 5- and 10-second durations. For 15-second clips, use WAN 2.6 image-to-video or join multiple WAN 2.5 clips in an editor.
What happens if my uploaded audio is longer than the video?
Only the first segment that matches the selected video duration is used. If the audio is shorter than the video, the rest of the clip plays silently.
Which languages does WAN 2.5 support?
It is tuned for Chinese and English and handles other languages with varying accuracy. Write the prompt and dialogue in the target language for the best chance of a clean result.
How do I get better lip sync?
Keep the face large in frame, use a near-frontal head angle, quote a short line, and upload a clean voice track if you have one. Avoid covered mouths, heavy shadows and long sentences.
Is WAN 2.5 better than WAN 2.6?
They solve different problems. WAN 2.5 is cheaper, supports 480p drafts and accepts uploaded audio. WAN 2.6 adds 15-second clips and 1080p at a higher per-second price.
Does WAN 2.5 support a negative prompt?
Yes. Unlike WAN 2.2 Spicy, this endpoint exposes a negative prompt and a seed, so you can suppress artifacts and reproduce a result.
Compare with other image-to-video models
- WAN 2.2 Image-to-Video
- WAN 2.6 Image-to-Video
- WAN 2.2 Spicy
- Seedance 2.5 Image-to-Video
- Kling O3 Pro Image-to-Video
- Browse all AI video models
Specifications
- Provider: Wavespeed
- Model type: Image to video
- Credit pricing: 480p: 20 credits · 720p: 30 credits · 1080p: 40 credits
- Capabilities: new, cinematic, starter, audio-sync, multi-shot
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.
Explore more AI tools
- Seedance 2.5 Text-to-Video
- Seedance 2.5 Turbo Text-to-Video
- Seedance 2.5 Image-to-Video
- Seedance 2.5 Turbo Image-to-Video
- Seedance 2.0 Mini Text-to-Video
- Seedance 2.0 Mini Image-to-Video
- MiniMax H3 Text-to-Video
- MiniMax H3 Image-to-Video
- MiniMax H3 Reference-to-Video
- Age Transform
- Virtual Try-On
- Photo Restore