Gemini Omni coming soonLearn more
Wan 2.7 Video Edit API: Instruction-Based Video Editing for Production Workflows
Tutorial

Wan 2.7 Video Edit API: Instruction-Based Video Editing for Production Workflows

EvoLink Team
EvoLink Team
Product Team
May 22, 2026
6 min read
Wan 2.7 Video Edit (wan2.7-video-edit) is the first Wan route exposed on EvoLink for editing existing videos through natural language instructions. Before this, every iteration on a generated clip meant re-generating the entire thing. Now you describe the change — "swap the background to a rainy street", "convert to anime style", "change the jacket to red" — and the model edits the clip while preserving the original motion.

This guide covers five real editing scenarios, the exact API parameters, cost math, and production patterns for teams shipping video editing into products.

For the full Wan 2.7 integration guide covering all four modes, see the Wan 2.7 API guide. For the model overview and playground, visit the Wan 2.7 model page.

TL;DR

  • Model ID: wan2.7-video-edit
  • Input: 1 source video (mp4/mov, 2-10 seconds) + text instruction + optional reference images (up to 4)
  • Output: Edited video at 720p or 1080p
  • Billing: Input video duration + output video duration. Failed tasks are free.
  • EvoLink pricing: $0.086/sec at 720p. A 5-second edit costs ~$0.86 (5s input + 5s output = 10s billed).
  • On EvoLink's current Wan routes, video editing is exposed through Wan 2.7. Wan 2.6 and 2.5 routes support generation only.

How it works

Source video + Text instruction + (optional) Reference images ↓ Wan 2.7 Video Edit API ↓ Edited video (preserves original motion)

The model preserves the structure and motion of the original clip. It changes the aspects you describe — style, colors, objects, backgrounds — without re-generating the motion from scratch. This makes iteration cycles dramatically cheaper than full re-generation.


API parameters

ParameterRequiredDefaultDescription
modelYeswan2.7-video-edit
promptYesNatural language edit instruction
video_urlsYesArray with exactly 1 source video (mp4/mov, 2-10 sec)
image_urlsNoUp to 4 reference images for style/content guidance
keep_original_soundNofalsetrue = preserve original audio; false = model handles audio
durationNo00 = keep original video length; explicit: 2-10 sec
qualityNo720p720p or 1080p
callback_urlNoHTTPS webhook for task completion

Scenario 1: Style transfer

Convert a real-world clip into an artistic style.

curl -X POST https://api.evolink.ai/v1/videos/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.7-video-edit",
    "prompt": "Convert the entire scene to a clay-style look with soft rounded edges and pastel colors",
    "video_urls": ["https://your-cdn.com/street-scene.mp4"],
    "keep_original_sound": true,
    "duration": 0
  }'
Use cases: Social media content variations, art direction testing, brand style exploration.

Scenario 2: Background replacement

Swap the background while keeping the foreground subject intact.

{
  "model": "wan2.7-video-edit",
  "prompt": "Replace the background with a rain-soaked Tokyo street at night, with neon reflections on wet pavement",
  "video_urls": ["https://your-cdn.com/person-walking.mp4"],
  "keep_original_sound": false,
  "duration": 0
}
Use cases: Product demos in different environments, virtual staging, ad localization.

Scenario 3: Object and clothing change

Change specific objects or clothing in the video using a reference image.

{
  "model": "wan2.7-video-edit",
  "prompt": "Replace the girl's outfit with the clothes from the reference image",
  "video_urls": ["https://your-cdn.com/fashion-clip.mp4"],
  "image_urls": ["https://your-cdn.com/target-outfit.png"]
}
Use cases: Fashion try-on, product placement, e-commerce video variants.

Scenario 4: Colorization

Convert black-and-white or desaturated footage to full color.

{
  "model": "wan2.7-video-edit",
  "prompt": "Convert this black-and-white footage to realistic color, maintaining the original lighting and shadows",
  "video_urls": ["https://your-cdn.com/vintage-footage.mp4"],
  "keep_original_sound": true,
  "duration": 0
}
Use cases: Archive restoration, documentary production, vintage content revitalization.

Scenario 5: Lighting and mood shift

Adjust the lighting and color grading without changing the content.

{
  "model": "wan2.7-video-edit",
  "prompt": "Shift the lighting to warm golden hour, with long shadows and soft lens flare",
  "video_urls": ["https://your-cdn.com/outdoor-scene.mp4"],
  "keep_original_sound": true,
  "duration": 0
}
Use cases: Color grading alternatives, mood board testing, time-of-day simulation.

Pricing and cost math

Video edit billing is based on input video duration + output video duration.
ExampleInputOutputBilled secondsCost (720p)
5-sec clip, keep original length5s5s10s$0.86
8-sec clip, keep original length8s8s16s$1.33
10-sec clip, trim to 5s output10s5s15s$1.25
Failed tasks are not billed. If the edit fails, you pay nothing.
For comparison, re-generating a clip with wan2.7-text-to-video at 5 seconds costs $0.43 — but you lose the original motion and there's no guarantee the new output matches. Video edit preserves the motion for roughly 2× the single-generation cost, which is almost always worth it for iteration workflows.

Production patterns

When to use video edit vs re-generation

ScenarioUse video editUse re-generation
Style change on approved motion
Background swap
Clothing/object replacement
Complete scene change
Motion is wrong
Starting from scratch
Rule of thumb: If the motion is right but the visuals need adjustment, use edit. If the motion itself needs to change, re-generate.

Prompt writing tips

  • Be specific. "Make it look better" fails. "Convert to watercolor painting style with visible brushstrokes" works.
  • Reference images help. When describing a target style or outfit, providing a reference image (image_urls) gives the model a concrete visual target.
  • Preserve audio when needed. If the original clip has dialogue or music you want to keep, set keep_original_sound: true.
  • Test with duration: 0 first. This keeps the original video length, which is the safest default for iteration.

FAQ

Can I chain multiple edits on the same clip?

Yes. Take the output of one edit and pass it as the input to the next. Each edit is an independent API call. Be aware that each pass adds quality degradation, so limit chains to 2-3 edits.

What video formats are supported?

mp4 and mov. The source video must be 2-10 seconds long.

Can I change the aspect ratio during editing?

The output will match the source video's aspect ratio by default. The model preserves the original composition.

Is this the same as Runway or Pika video editing?

Wan 2.7 Video Edit is an API — it doesn't have a visual editor UI. It's designed for developers building video editing into products, not for end-user creative tools. The advantage is programmatic control and batch processing at $0.086/sec.


Next steps

Ready to Reduce Your AI Costs by 89%?

Start using EvoLink today and experience the power of intelligent API routing.