WAN 2.2 Video Edit
Edit footage with a text prompt using WAN 2.2 Video Edit. Clips from 3–60s at 480p or 720p, with identity, outfit and style changes from 30 credits/s.
WAN 2.2 Video Edit changes existing footage from a text instruction. Describe the edit — swap a character, change a garment, recolour a scene, add a prop — and the model applies it while keeping the original motion and continuity. It handles clips from 3 to 60 seconds at 480p or 720p, priced at 30 and 50 credits per second respectively.
- Input: Video + prompt
- Duration: 3s – 60s
- Price: 30–50 credits/s
- Resolution: 480p / 720p
How to use WAN 2.2 Video Edit
One instruction per run. The model preserves motion best when it only has to change one thing.
- Trim the source to the shot you need Upload only the footage that should change. Every second is billed, and extra context gives the model more to disturb.
- Write a single, concrete edit Name the object and the change: replace the red jacket with a black coat, add rain, remove the logo. Multiple changes in one prompt tend to half-apply.
- Pick the resolution 480p is the default and the cheaper test. Use 720p when faces, text or fine garment detail must survive.
- Check continuity frame by frame Watch for flicker or identity drift at the edit boundary. If the change pulses between frames, rerun with a simpler instruction.
- Lock the seed for a series When applying the same edit across several shots, reuse the seed and the prompt so the look stays consistent.
Settings reference
WAN 2.2 Video Edit applies a text instruction to existing footage while keeping the original motion and continuity.
| Setting | Value | Notes |
|---|---|---|
| Source video | Required, up to 60s | The edit is applied to this footage; output follows its length and motion. |
| Edit prompt | Required | Describe one change: swap a garment, change the weather, add a prop, restyle a scene. |
| Resolution | 480p / 720p | 30 credits per second at 480p, 50 at 720p. 480p is the default. |
| Seed | Optional | Reuse a seed to keep an edit consistent across a series of shots. |
| Billing | 3s – 60s | Billed per second of source footage, with a 3-second minimum. |
| Duration | Follows the source | There is no duration selector; trim the source before uploading. |
What text-driven editing can change
The model is built around attribute changes rather than structural ones. Identity replacement, clothing and prop swaps, colour changes and broad style shifts are within scope; changing the camera move, the blocking or the number of people in a shot is not. It works because the original frames supply timing and composition while the prompt supplies the new attribute.
Continuity is the main advantage over regenerating a shot. If a performance is good but the wardrobe is wrong, an edit preserves the performance. It also means the result inherits any problems in the source: shaky footage stays shaky, and soft focus stays soft.
- Identity and character replacement
- Outfit, prop and hair changes
- Colour and style adjustments
- Keeps original timing and camera
- Does not restructure the shot
Pricing by clip length
Billing runs on the uploaded clip's length, from a 3-second minimum to a 60-second maximum. 480p costs 30 credits per second and 720p costs 50. A 5-second 480p edit is 150 credits; a 30-second 720p edit is 1,500.
The 60-second ceiling is a product limit for cost control; the underlying model can accept more, but single runs beyond 60 seconds are not offered here. Split longer footage into shots and edit each one, which also makes prompt mistakes cheaper to isolate.
- 30 credits per second at 480p
- 50 credits per second at 720p
- Minimum billed length: 3 seconds
- Maximum: 60 seconds
- No per-run fee
| Clip length | 480p | 720p |
|---|---|---|
| 3s | 90 | 150 |
| 5s | 150 | 250 |
| 10s | 300 | 500 |
| 30s | 900 | 1,500 |
| 60s | 1,800 | 3,000 |
Prompt patterns for clean edits
Use a change-plus-constraint sentence. Name the subject by position, state the new attribute and list what must remain. Change the woman's vest to a red dress while keeping her face, motion and the background unchanged is clearer than make her outfit red, which leaves the model to decide what else may change.
For identity replacement, describe the replacement rather than the original: replace the man with a teenage boy, keeping the same walking pace and camera angle. For colour and style, name the target palette or medium instead of a vague mood. If the output overcorrects, narrow the prompt before increasing the strength — there is no strength slider on this endpoint.
- Change X to Y while keeping Z
- Identify subjects by position
- Name the new attribute precisely
- List what must stay unchanged
- One main edit per pass
Working with longer clips
A 60-second edit is expensive enough that you should test the prompt on a shorter section first. Cut a 5–10 second representative segment, run the edit, and confirm the wording before applying it to the full clip. This also reveals whether the model can see the detail you are referring to.
Keep the source at the same resolution you plan to output. Editing a 720p clip at 480p saves credits but throws away detail; editing a 480p clip at 720p adds no real information and may amplify compression artefacts. For final delivery, edit at the source's native tier.
- Test on a short representative section
- Apply the winning prompt to the full clip
- Match output resolution to source
- Split long footage into shots
- Review at full speed for temporal artefacts
Limits and alternatives
The endpoint stops at 720p and 60 seconds and has no audio controls. Fine text, logos, hands and fast occlusion are weak points because frames are redrawn. It is not a compositor and should not be used where a label or legal text must remain pixel-identical.
HappyHorse 1.0 Video Edit is the higher-resolution option: 720p or 1080p, 3–15 seconds, with up to nine reference images. Kling Video Edit and Kling O3 Pro Video Edit offer different speed and quality trade-offs at shorter durations. WAN 2.2 Video Edit is the choice when the clip is longer than 15 seconds and 720p is enough.
- 720p maximum
- 60-second maximum
- No audio controls
- HappyHorse edit for 1080p up to 15s
- Kling edits for shorter premium passes
Prompt examples for WAN 2.2 Video Edit
- Outfit swap: Change the woman's vest to a red dress while keeping her face, hair, walking motion, camera angle and the street background unchanged. State what must stay fixed.
- Identity replacement: Replace the man with a teenage boy, keeping the same walking pace, body position, camera tracking speed and lighting. Describe the replacement's build.
- Colour grade: Turn the warm afternoon scene into cool blue evening light while keeping the subject's motion and framing identical. Name the target palette.
- Prop change: Replace the paper cup in the woman's hand with a glass bottle, keeping her hand position and stride unchanged. Name the hand and object.
- Background restyle: Change the plain office background into a rain-soaked neon street while keeping the subject and camera move unchanged. Background edits may affect edge lighting.
- Style transfer: Restyle the footage as a 1980s anime scene while preserving the original timing, silhouettes and camera path. Broad style edits tolerate more drift.
Pro tips for WAN 2.2 Video Edit
This is a targeted editor, not a scene generator. Small, legible changes survive; whole-scene rewrites do not.
- One edit per pass Need a new garment and new weather? Run two edits in sequence. A combined prompt forces the model to trade one change against the other.
- Keep the camera language The source motion is preserved, so do not describe camera moves in the prompt. Describe only what changes inside the frame.
- Trim before uploading A 60-second source at 720p costs 3,000 credits. Uploading only the ten seconds that matter cuts the bill by six times.
- 720p for text and faces Logos, lettering and facial detail smear at 480p. Pay for 720p when those elements are the point of the edit.
- Seed for consistency across shots Reusing a seed across a series keeps color and texture decisions aligned, which matters when the shots will be cut together.
Frequently asked questions
How much does WAN 2.2 Video Edit cost?
30 credits per second at 480p and 50 credits per second at 720p, billed on the input clip length from 3 to 60 seconds.
How long can the input be?
From 3 to 60 seconds. Videos under 3 seconds are billed at 3 seconds.
What resolutions are supported?
480p and 720p.
Does it keep the original audio?
Audio is not part of the edit controls. The endpoint returns the edited visual result; re-attach the original track in post if needed.
Can it change the camera move or add people?
No. It changes attributes within existing footage. Structural changes such as camera moves, new blocking or extra characters are outside its scope.
What edits work best?
Identity replacement, outfit and prop changes, colour grades and broad style shifts. Fine text, logos and hands are less reliable.
Can I edit a 60-second clip in one run?
Yes, but test the prompt on a short section first. A full 60-second 720p edit costs 3,000 credits.
How does it compare with HappyHorse Video Edit?
HappyHorse offers 1080p and up to nine reference images but only 3–15 seconds. WAN 2.2 reaches 60 seconds at 480p or 720p.
Compare with other video editing models
- HappyHorse 1.0 Video Edit
- Kling Video Edit
- Kling Video Edit Fast
- Kling O3 Pro Video Edit
- WAN 2.2 Animate
- Browse all AI video models
Specifications
- Provider: Wavespeed
- Model type: Video editing
- Credit pricing: 480p: 30 credits · 720p: 50 credits
- Capabilities: popular, cinematic, realistic, editing
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.