跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
视频与音频

ugc-editing

Covers assembling raw footage, trimming bad takes and silences, choosing and applying a talking-head/b-roll format (straight intercut, stacked split, or masked/floating overlay), and caption placement for UGC-style content — agnostic to whether the clips are AI-generated or real filmed footage. Use whenever the user wants to edit, cut, assemble, format, or caption UGC-style video, or pick a talking-head + b-roll layout. Picks up after ugc-photo-prompts/ugc-video-prompts if the footage is AI-generated, or after real footage is imported via import_media.

palmier-io · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add palmier-io/palmier-skills --skill ugc-editing

技能说明


name: ugc-editing description: Covers assembling raw footage, trimming bad takes and silences, choosing and applying a talking-head/b-roll format (straight intercut, stacked split, or masked/floating overlay), and caption placement for UGC-style content — agnostic to whether the clips are AI-generated or real filmed footage. Use whenever the user wants to edit, cut, assemble, format, or caption UGC-style video, or pick a talking-head + b-roll layout. Picks up after ugc-photo-prompts/ugc-video-prompts if the footage is AI-generated, or after real footage is imported via import_media.

UGC Editing

Core principle

Editing is footage-source-agnostic. The same assemble → trim → layout → caption pipeline applies whether the clips came from ugc-video-prompts or a real phone shoot. This skill picks up wherever footage already exists and turns it into a finished vertical post. The single biggest editing mistake in UGC is padding — long pauses, repeated stumbles, slow setups. Cut harder than feels comfortable. Authentically raw ≠ unedited; it means tight pacing with natural-sounding joins.

Tool model (current Palmier): timeline positions are project frames; source trims are seconds via source: [start, end]. Linked A/V is folded into the video clip as audio: { id, … } — mute/edit audio by that nested id. Mutations return a timeline delta — patch from it; don't re-get_timeline after every call. Index 0 renders on top.


Step 0: Pre-flight — always run before touching the timeline

  1. get_timeline — check resolution / canGenerate. Must be 9:16 (typically 1080×1920). If not:

    set_project_settings({ aspectRatio: "9:16", quality: "1080p" })
    

    Do this before placing clips. Explicitly configured settings stick (the first clip no longer snaps the project to its source aspect — placement may only note a mismatch).

  2. get_media — identify A-roll and b-roll. Note durationSeconds, width/height, hasAudio.

  3. inspect_media on the A-roll:

    • First: overview: true — storyboard + segment timestamps.
    • Then: wordTimestamps: true (window long clips with startSeconds/endSeconds) — read words as prose before cutting. Look for retakes, fillers, false starts, trailing dead air, weak sections.

Step 1: Place A-roll only — no b-roll yet

add_clips({
  entries: [{
    mediaRef: AROLL_ID,
    startFrame: 0,
    // whole asset, or trim in SOURCE seconds (not project frames):
    source: [SKIP_PREROLL_S, END_S]   // omit for full file
  }]
})
// omit trackIndex on every entry → auto-creates video + linked audio tracks
  • Never pass trackIndex here when you want a fresh storyline track. Omitting it auto-creates one shared video track (and linked audio). Specifying an existing index can overwrite whatever is already there.
  • There is no durationFrames / trimStartFrame on add_clips — use source: [startSeconds, endSeconds] for source span, or endFrame (mutually exclusive with source) for an exact timeline occupy range.
  • 16:9 A-roll in a 9:16 project may letterbox. Cover-crop the talking head to full frame:
    apply_layout({
      layout: "full",
      slots: [{ slot: "main", clipIds: [AROLL_CLIP_ID], anchor: "center" }]
    })
    
    Prefer apply_layout over set_clip_properties transforms — layout sets transform and crop together (transform-only leaves a stale crop and can stretch).

Step 2: Cut bad takes and filler — words first

Prefer remove_silence first for dead air (no transcript), then word cuts.

Read first, cut once per pass. get_transcript → mark indices → one remove_words call. Indices shift after every cut — re-read before the next pass.

get_transcript()
remove_words({
  words: [
    [FIRST_RETAKE_START, FIRST_RETAKE_END],
    FILLER_INDEX,
    [FALSE_START_START, FALSE_START_END],
  ],
  cutAggressiveness: "tight"   // "balanced" / "loose" if joins feel clipped
})
// bulk fillers: remove_words({ matches: ["um", "uh", "hmm"] })

What to cut: duplicate retakes (keep the most fluent), fillers, false starts, trailing soft endings, off-topic tangents that break hook → value → CTA.

What to keep: natural pauses between real thoughts; best take of each section; the outro/CTA.

Captions block ripple: if a caption track exists, remove it before cutting:

manage_tracks({ remove: [{ trackId: CAPTION_TRACK_ID }] })
// or remove: [index] — prefer trackId from get_timeline

Then cut, then re-add captions. Never caption before cutting is finished. There is no remove_tracks tool.


Step 3: Pick a layout format

Format 0 — Straight intercut (default)
B-roll is supplementary, not simultaneous. Hard cuts alternating full-frame A-roll and b-roll on the same video track every 3–5s. Use add_clips (overwrite landing region) or insert_clips (ripple — requires trackIndex) — no layout tool.

Format 1 — Stacked split, b-roll top / talking head bottom
B-roll is proof (product/result/location). Viewer needs both at once. layout: "top_bottom", b-roll → top, head → bottom.

Format 2 — Stacked split, head top / b-roll bottom
B-roll is filler/retention. Same layout, slots flipped.

Format 3 — Full-bleed b-roll, floating head
Not clean via apply_layout alone — needs chroma/apply_effect + manual framing. Skip unless asked.


Step 4: Place b-roll

Format 0 — intercut on the storyline track

add_clips({
  entries: [
    { mediaRef: BROLL_1, startFrame: F0, source: [0, 4] },
    { mediaRef: BROLL_2, startFrame: F1, source: [0, 3] },
  ]
})
// same track as A-roll → overwrites those ranges (split/trim under the hood)

Mute any linked b-roll audio (nested audio.id on each placed video, or mute the b-roll audio track via manage_tracks).

Format 1–2 — stacked split (critical: span-scope the head)

apply_layout on an existing clip is whole-clip (time-invariant). If you layout the entire A-roll

阅读完整来源 ↗