Edit Video

Change an existing clip with a prompt. The source keeps the motion, timing, and framing. The prompt says what to change and what to keep.

Inputs to collect

• Source clip: a public https .mp4 URL or a TokenLab result URL. Know its length and resolution (ffprobe).
• The change: a style ("watercolor animation"), a scene change ("make it night, wet streets"), or an object change ("the jacket is red leather").
• What must stay: pose, motion, framing, faces, product shape.
• Reference images (optional): a style or object to copy, on models that take them.

Models

<!-- tokenlab:models:start --> <!-- Generated from the live catalog by scripts/sync_models.py. Do not edit; change preferredModels in catalog.json. -->

Operation	Model	Pick	Use
video-to-video	grok-imagine-video	default	Prompt-only restyle of a short clip
video-to-video	kling-omni-video	alternative	Change guided by reference images (not live-tested)
video-to-video	flux-video-edit	alternative	Prompt-only edit, alternative
video-to-video	happyhorse-1.0	alternative	References, alternative (not live-tested)
video-to-video	seedance-2.0	alternative	Alternative (not live-tested)

This list refreshes daily from the live catalog. Preferred models come first; otherwise models are ranked by their 24-hour success rate. Before a call, confirm the model with GET /v1/models/{model} and read its price from GET /v1/models/{model}/pricing.

<!-- tokenlab:models:end -->

How to choose

Exact fields and values come from each model's contract.

• Grok Imagine video: a prompt-only restyle of a short clip. Output is capped at 8.7 s. No duration, resolution, or aspect_ratio.
• Kling Omni video: a change guided by reference images.
• FLUX video edit: a prompt-only alternative. Source up to 15 s, 720p, and 50 MiB. duration is required.
• HappyHorse: an alternative that takes reference images. Output is capped at 15 s, at 720p or 1080p.
• Seedance 2.0: an alternative.

All use POST /v1/videos/generations with operation: "video-to-video". Seedance may reject a source clip that shows a realistic person.

Workflow

Run each call with tokenlab-run (contract → estimate → call → poll → result). Commands below use its helper, ../tokenlab-run/scripts/tokenlab_run.py, a path relative to this skill's folder. If you installed this skill without tokenlab-run, send the same requests with curl. --wait stops after 100 s; a longer task keeps running, so continue with the resume command it prints.

1. Read the contract: python ../tokenlab-run/scripts/tokenlab_run.py contract MODEL --operation video-to-video. Note supported_parameters, notes, and delivery. Send only listed fields. Grok Imagine video does not use duration, resolution, or aspect_ratio for this operation; leave them out.

2. Check the source against the model's limits (Grok Imagine video: 8.7 s output cap; FLUX video edit: 15 s, 720p, 50 MiB; HappyHorse: 15 s output cap). Trim the clip to the shot you need first.

3. Estimate the cost: python ../tokenlab-run/scripts/tokenlab_run.py price MODEL, then per-second price × output seconds. On these models the output length follows the source. Grok Imagine video holds the cost of its 8.7 s maximum up front and charges the delivered length. Use the Official price as the upper bound. Tell the user the estimate.

4. Build the body. A prompt-only model with no length field (Grok Imagine video):

   {"model": "MODEL_ID", "operation": "video-to-video", "video_url": "https://....mp4", "prompt": "Turn the scene into a hand-drawn watercolor animation. Keep her pose, motion, and framing."}
   

   A model that requires duration (FLUX video edit):

   {"model": "MODEL_ID", "operation": "video-to-video", "video_url": "https://....mp4", "prompt": "...", "duration": 2}
   

   Take MODEL_ID from the Models list above for video-to-video, and send only the fields its contract lists.

5. Call and wait: python ../tokenlab-run/scripts/tokenlab_run.py run /v1/videos/generations --json @body.json --wait. Save the task ID. Poll every 5-10 s.

6. If a wait times out, resume with python ../tokenlab-run/scripts/tokenlab_run.py poll TASK_ID. Never create a duplicate task.

7. Compare the result with the source (first frame, last frame, and one middle frame). Return the video URL, the model, the task ID, and the estimate.

Technique

• Say what changes and what stays. "Change X to Y. Keep the pose, camera move, and framing." Models keep more of the source when you name it.
• One change per pass. Style first, then an object change in a second pass. Two big changes at once drift more.
• Style words that travel well: a medium ("watercolor", "claymation", "pencil sketch", "anime cel shading"), a palette ("muted teal and orange"), and a light change ("golden hour", "neon night").
• Short, clean sources. Trim to the shot you need. Hard cuts inside the source often break the edit.
• Source audio. Edits can keep the source soundtrack. Do not expect the edit to remove or replace sound; use an editor for audio.
• Duration on FLUX video edit. Send duration equal to the source length in seconds. It sets the first hold; the final charge follows the delivered clip.
• Reference images. On Kling Omni video and HappyHorse, add a reference image for a precise look or object instead of a long description.

Quality bar

• Motion, timing, and framing match the source.
• The requested change shows in every frame, with no flicker between styles.
• Faces, hands, and products keep their shape; nothing the user asked to keep has changed.
• The output length matches the source (within the model's cap); the user saw the estimate first.

Related skills

• tokenlab-run: the call, poll, and result mechanics.
• tokenlab-extend-video: continue the clip instead of changing it.
• tokenlab-motion-control: put a new character into this clip's motion.
• tokenlab-edit-image: edit a single frame or still.