SuperAPI AI Directory

Video Editing

Design & creation

View repository

Updated 11-01 21:05

About this skill

Video Editing

Mission

Transform an existing video into an intentional child asset: edit its visual content, extend its ending, create a first/last-frame transition, or combine approved clips. Preserve the source relationship, audio decision, and exact requested change so the user can compare parent and child outputs.

Use Short-form Repurposing for highlight selection, autocrop, and caption packaging. Use Video Enhancement only after a parent is selected.

Required inputs

  • source video URL or local file and its role;
  • requested edit, extension, transition, or assembly outcome;
  • target duration, ratio, resolution, and delivery format;
  • preservation list: subject, scene, audio, dialogue, music, timing, or text;
  • optional reference images, last frame, audio, or prompt; and
  • rights/consent state and approval boundary.

Ask whether the original audio should be preserved, replaced, mixed, or removed. Do not infer this from the visual edit request.

Choose the operation

Request Operation Important evidence
Change or restyle content in an existing clip video-to-video edit parent URL, edit prompt, preserved regions
Continue a clip video extension parent URL, continuation intent, endpoint duration
Transition between two visual endpoints first/last-frame or model-specific generation ordered first and last images
Join approved clips combine utility ordered clip IDs, transition/audio policy
Change framing for a selected segment reframe utility use the repurposing workflow

Resolve exact endpoint fields from MODELS.md and the live catalog. Current families expose different shapes: Seedance 2.5 has video-edit and video-extend paths; Wan 2.7 has edit/extend paths; Veo 3.1 extension uses a provider request ID and prompt; and first/last-frame paths require an ordered image list. These are not interchangeable aliases.

Execute

  1. Assign a stable parent asset ID and inspect its duration, dimensions, audio tracks, and visible text if those checks are available.
  2. Upload a local source or reference via media.upload_file; retain the returned hosted URL and expiry expectation. Do not put private media in a public prompt or receipt.
  3. Check the live schema and estimate cost when relevant. Submit only accepted fields. For extension, retain both the provider request ID and the job prediction ID when the endpoint uses both.
  4. Poll or receive the result, preserving status and provider error. Do not silently fall back from edit to generation: the output semantics differ.
  5. Review parent versus child for requested change, unwanted drift, temporal artifacts, cut points, audio continuity, captions, and export properties.
  6. Combine only approved child assets in explicit order. State whether audio was retained, replaced, ducked, or omitted.

Use video QA for the review gate. If the edit cannot preserve an invariant, return the failure and a smaller alternative such as a conventional post-production edit or a new reference-led shot.

Output contract

Return a parent/child table with operation, model/endpoint, input URLs or IDs, prediction ID, requested fields, output URL, status, audio treatment, QA notes, and next action. Include all failed branches and do not overwrite the parent. Save a receipt under .video/ only when permitted, following reports-and-receipts.md.

Editing is draft-only. External publishing and platform changes are outside this skill.