Extend Video

Add new footage after the last frame of a clip. There are two kinds of input: a public video URL (any clip that meets the size and content rules), or the task ID of a clip that the same model made on TokenLab.

Inputs to collect

• Source clip: a public https .mp4 URL, or a TokenLab result URL. If the clip was made on TokenLab, also keep its task ID and model.
• What happens next: the action and camera for the new part. Keep the same subject, place, and light.
• Added length and output resolution.
• Sound: with or without audio.

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-extension	seedance-2.0	default	Continue any public clip from its video_url
video-extension	seedance-2.0-fast	alternative	Faster, up to 720p (not live-tested)
video-extension	seedance-2.5	alternative	Newer tier, up to 720p (not live-tested)
video-extension	seedance-2.0-mini	alternative	Low-cost, up to 720p
video-extension	veo3.1-fast	alternative	Known limit: task_id extension of a clip this model made returned 400 in live tests; prefer Seedance
video-extension	grok-imagine-video	alternative	Known limit: task_id extension of a clip this model made returned 400 in live tests; prefer Seedance

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.

• Seedance 2.0: the default. Continues any public clip from its video_url. Adds 4-15 s, 480p up to 4k.
• Seedance 2.0 Fast: faster, up to 720p.
• Seedance 2.5: the newer tier, up to 720p. Adds up to 30 s.
• Seedance 2.0 Mini: low cost, up to 720p.
• Veo 3.1 Fast and Grok Imagine video: extend only a clip that the same model made, by task_id. Known limit: every live test of this path returned 400, so treat it as unreliable and prefer Seedance.

All use POST /v1/videos/generations with operation: "video-extension". Veo and Grok extension have Verified supply only, and they need the task ID of a clip from the same model. If TokenLab rejects the task ID with a 400 (for example record not found or cannot currently honor these request parameters), that clip cannot be extended on that model. Use the Seedance path with the clip's video_url.

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-extension. Note the input fields (video_url or task_id), durations, resolutions, and delivery.

2. Check the source clip (Seedance path): it shows no realistic human face, and each frame has at least 407,696 pixels (for example 640×640 or 1280×720). Clips that break either rule fail with 400 when you create the task. Use ffprobe to read the size.

3. Estimate the cost: python ../tokenlab-run/scripts/tokenlab_run.py price MODEL.

   • Seedance prices per million video tokens, with a separate price when the request has video input. The task result reports usage.total_tokens; the charge is tokens × that price. For scale, a 4 s 480p extension of a 2 s 720×720 clip reported 67,600 tokens.
   • Per-second models: per-second price × added seconds. Use the Official price as the upper bound. Tell the user the estimate.

4. Build the body:

   {"model": "MODEL_ID", "operation": "video-extension", "video_url": "https://....mp4", "prompt": "Extend the video: the steam keeps rising from the full mug, and the camera slowly pushes in. Same morning window light.", "duration": 4, "resolution": "480p", "output_audio": false}
   

   Take a Seedance MODEL_ID (seedance-2*) from the Models list above for video-extension. Set duration and resolution to values 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. The Seedance result is the new part only, starting from the source's last frame. Join the source and the new part in an editor or with ffmpeg (see Technique).

8. Return the new video URL, the model, the task ID, the estimate, and the joined file if you made one.

Technique

• Continue, do not restart. Start the prompt with the intent ("Extend the video: ...") and describe only the next action. Repeat the key look words from the original prompt (subject, place, light) so the style holds.
• One beat per extension. Add one action per call. For a long scene, extend several times and check each part before the next call.
• Match the source settings. Use the same aspect ratio and resolution as the source, so the parts join cleanly.
• Seedance input rules. Seedance rejects a source clip that "may contain real person", and that check also caught a photoreal AI-generated person. Real people need TokenLab's Seedance material and real-person verification flow. Very small clips are rejected for pixel count.
• Grok fields. Grok Imagine video takes task_id, prompt, extend_at (the second in the source where the new part starts), and extend_times (the added seconds as a string, for example "6"). It does not take duration or resolution for this operation.
• Veo fields. Veo 3.1 and Veo 3.1 Fast take task_id (or origin_task_id) and prompt. Google's Veo docs say an extension continues from the final second of the clip and that a voice is hard to continue if it is not present in the last second.
• Join the parts. When both files have the same size, frame rate, and audio layout: ffmpeg -i part1.mp4 -i part2.mp4 -filter_complex "[0:v][1:v]concat=n=2:v=1:a=0[v]" -map "[v]" joined.mp4 (video only). Add audio back in the editor if you need it.

Quality bar

• The first frame of the new part continues the last frame of the source: same subject, place, light, and framing.
• No jump in color, size, or position at the join.
• The new action matches the prompt, and the added length matches the request.
• The user saw the estimate before the job, and the answer names the model and task ID.

Related skills

• tokenlab-run: the call, poll, and result mechanics.
• tokenlab-animate-image: start a new clip from a still.
• tokenlab-keyframe-transition: end on a chosen image.
• tokenlab-edit-video: change the existing clip instead of continuing it.