SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsyh-chatcut-edit

yh-chatcut-edit

General & other

View repository

Updated 10-16 09:35

About this skill

yh.edit for ChatCut

Core principles

  • Execute only when ChatCut is installed and callable, and an editable project exists; if ChatCut capability, a project, or the main asset is missing, state the missing items and stop.
  • Use ChatCut to create a project that can continue to be edited; do not bake it into a non-editable final video using external scripts.
  • Complete A-Roll first, then forcibly pause; continue to B-Roll only after the user explicitly confirms.
  • Use the current project, timeline, asset library, and the asset names provided by the user as the source of truth; do not guess asset content or corresponding positions.
  • After each modification, verify the project status and key frames. Any tool error, missing asset, timeline conflict, or uncertain result must be stopped and reported immediately; do not automatically proceed to the next stage.
  • Preserve the user's existing edits and unrelated tracks; do not clear the timeline, and do not overwrite modifications the user has not authorized.

Workflow state machine

Advance strictly according to the following states:

准备 -> A-Roll -> 等待确认 -> B-Roll -> 字幕 -> 音效 -> 背景音乐 -> 总检

Do not skip "awaiting confirmation," and do not put all stages into one continuous execution.

0. Preparation

  1. Read the ChatCut project, main timeline, asset library, and existing tracks.
  2. Confirm the main talking-head asset, verbatim transcript, aspect ratio, frame rate, and the B-Roll assets and pairing relationships specified by the user.
  3. Load and follow chatcut:chatcut-plugin-basicsfirst; load chatcut:talking-head-guide、chatcut:asset-import、chatcut:transcription、chatcut:voice、chatcut:music and chatcut:verification。
  4. If assets or mappings are incomplete, first do the preparation work that does not depend on the missing information, then ask the minimum necessary questions.

1. A-Roll editing

  1. Process only the main talking head and its person footage: remove obvious pauses, repetitions, slips of the tongue, stumbles, and empty leading/trailing gaps.
  2. Maintain the original meaning and narrative order; do not delete arguments, turns, or necessary breaths just to pursue brevity.
  3. Check each cut point to avoid swallowed words, broken words, abrupt jumps in audio, and unnatural half-sentences.
  4. By default, adjust all valid talking-head segments to 1.1x; if the user explicitly requests another speed, the user's request prevails. Keep the pitch natural, and make all subsequent work based on the timeline of this final pacing.
  5. When the main talking head is person footage, moderately enlarge the person footage (Punch-in) at key semantic points such as the hook, the landing of a point, numerical amounts, reversals, or emotional emphasis, to strengthen key points and visual rhythm; if the user explicitly requests keeping the frame unchanged or locking the A-Roll, do not execute.
  6. Normal shot sizes keep the current base ratio (usually 100%when unadjusted); key shot sizes are enlarged by 5%–15%on top of the base ratio in principle; if the asset clarity is insufficient, reduce the enlargement amount, and do not cause obvious blurriness by pursuing close-ups.
  7. Enlargement should cut in at the beginning of the key sentence or at the stressed point, and return to the normal shot size when the semantics end; prefer clean shot-size changes, avoid arbitrary changes in the middle of a half-sentence, continuous push-pull, or mechanical zooming on every sentence. Keep a normal shot size between two adjacent enlargements so that the key points form contrast.
  8. After enlargement, check the person's eyes, top of head, gestures, and subtitle safe area to avoid incorrect cropping; shot-size changes may be used to soften jump cuts, but must not be used to conceal swallowed words, broken words, or unnatural audio cut points.
  9. Do not add B-Roll, subtitles, sound effects, background music, or MG animation.
  10. Listen through from beginning to end and spot-check cut points and shot size changes to confirm the spoken audio is coherent, has no repetition, has no obvious gaps, and that punch-ins correspond accurately to vocal emphasis.

A-Roll confirmation checkpoint (must pause)

After A-Roll is complete, you must:

  1. Briefly report what was completed, the final video duration, disputed points that were deleted or kept, the applied speech speed, and the punch-in positions and scale ranges used.
  2. Let the user preview first, and explicitly ask whether they confirm the A-Roll and whether you may proceed to B-Roll.
  3. End the current execution immediately. Do not continue in the same round to add B-Roll, subtitles, sound effects, or music.

Only proceed to the next stage after receiving an explicit reply such as "confirmed," "A-Roll is fine," or "continue with B-Roll." Ambiguous replies, additional comments, or error reports do not count as confirmation; fix the A-Roll first, then wait for confirmation again.

2. B-Roll

  1. First read the user-specified assets and place them according to the verbatim transcript's semantics and the correspondences given by the user.
  2. When the user explicitly specifies that a certain file be used for a certain sentence, strictly use that file and do not substitute it on your own.
  3. Keep the A-Roll original audio and use B-Roll to cover the visuals; do not let B-Roll's own audio interfere with the spoken audio unless the user explicitly requests it.
  4. B-Roll in-points follow keywords, and out-points follow the end of the semantic unit; avoid meaningless frequent switching and long static images.
  5. When image aspect ratios do not match, prioritize keeping the core information fully visible; it is acceptable to crop irrelevant bottom areas or leave designed margins, but do not crop out titles, the product subject, or key text.
  6. Only when assets are insufficient and the user permits it, supplement with generic stock images, collages, or minimalist MG; by default, do not generate decorative assets unrelated to the product.
  7. After completion, check segment by segment the asset names, spoken semantics, time ranges, and visual safe areas.

3. Subtitles

  1. Generate subtitles based on the final 1.1x speed timeline, using Simplified Chinese.
  2. Split cards by semantic short phrases, not by mechanically truncating at a fixed character count; "AI 剪辑 Agent," product names, amounts, and complete phrases must not be split across two cards.
  3. By default, remove Chinese commas and English commas, keeping question marks, exclamation marks, periods, and numeric symbols only when truly necessary.
  4. Prefer one to two lines per card to ensure complete display at large font sizes; when line breaks are truly needed, break at natural semantic boundaries.
  5. Check proper nouns, numbers, English capitalization, and each card's in-point and out-point.

4. Emphasis sound effects (optional)

  1. Add sound effects only at hooks, viewpoint landing points, numeric amounts, reversals, asset cut-ins, or important transitions.
  2. Prefer lightweight whoosh, pop, click, hit, cash, or riser sound effects; the choice should serve the semantics, and do not add them to every sentence.
  3. Control volume and duration so they do not cover the human voice; avoid stacking similar sound effects consecutively.
  4. Audition all sound effects one by one and check whether they are synchronized with the visuals and emphasis.

5. Background music

  1. Choose background music with a light, modern rhythm that does not overly steal the scene, consistent with the spoken topic and the visual tone.
  2. Lay it from the beginning to the end, looping or trimming as necessary; avoid abrupt breakpoints.
  3. Keep the music steadily under the human voice, with slight dynamic changes at hooks, turns, and the ending; it must not affect the clarity of the spoken audio.
  4. If music needs to be generated, use chatcut:music; if a suitable track already exists, prioritize reusing the project asset library.

6. Final review

  1. Play the complete timeline from start to finish, checking voiceover coherence, B-Roll mapping, subtitle semantics, sound effect synchronization, and music volume.
  2. Perform a visual verification at the beginning, A-Roll cut points, Punch-in in/out points, B-Roll in/out points, long subtitle sentences, sound effect accents, and the ending.
  3. Confirm there are no black frames, overlapping subtitles, offline assets, audio clipping, incorrect aspect ratios, or remnants at the end of the timeline.
  4. Report what was actually completed and the items that still require the user's judgment; do not describe unverified results as complete.

Execution boundaries

  • When the user requests only one stage, execute only that stage and do not automatically complete the entire workflow.
  • If the user requests changes before the confirmation gate, remain in the A-Roll stage, and after completing the changes, stop again and wait for confirmation.
  • After the user confirms the A-Roll, you may continuously complete the B-Roll, subtitles, sound effects, background music, and final review; if the user requests stage-by-stage confirmation, pause at each stage and wait for confirmation.
  • Export is not a default step. Only when the user explicitly requests export, load and use chatcut:export。