Clip Editorial Director
Follow this procedure whenever building a montage edit plan or refining hooks/titles for gta6_clips or wow_forever_clips.
Required Inputs
- Read
<project>/config/channel.json - Read
<project>/config/style_guide.md - Read
<project>/config/few_shot_plans.jsonand<project>/config/learnings.md - Read
clip_factory/shared_assets/sound_library.json(andmeme_map.json) - Read the user's curated clip list and directorial notes from
<project>/workspace/02_curated/
Editorial & v7 Production Rules
- Creator Rewards Target (62–85s Flexible): Target 62–75s for TikTok Creator Rewards, and allow up to ~85s when a clip's comedic setup/payoff genuinely needs room to breathe. Never pad with dead air.
- Dual-Zone Color-Coded Karaoke Captions:
- Diarize speech explicitly into
streamervs.npc— never dump all Whisper output intostreamer. streamer: Top zone near mouth (streamerCaptionY, default27%), white base with bright yellow (#FFE600) word-by-word karaoke highlight (wordsarray).npc: Bottom zone (bottom: 215px), italic electric cyan (#00F0FF) in quotes.editorial: Deadpan hot-pink mid-screen popups ("INSTANT REGRET","DOWN GOES GRANDMA").- Clamp any abnormally stretched Whisper words (
>1.35s) so captions never linger across silence.
- Diarize speech explicitly into
- Varied Camera Choreography & Clean Framing:
- Vary zoom modes per situation (
aggressive_punchup to2.05x+ImpactShakefor brutal hits;slow_creep+ FFmpegplaybackRateslow-mo for dramatic jaw-drop reveals;snap_holdfor punchy beats;smoothfor subtle focus). - Use
letterboxCrop: { yPct, hPct }on letterboxed 16:9 / 4:3 clips with black bars or baked top text so gameplay sits cleanly over a blurred darkened1080×1920backdrop.
- Vary zoom modes per situation (
- Copyright & Vocal Clarity Check:
- For
gta6_clips, check for car radio or unfitting clip music; setisolateVocals: truewhen swapping inbgmFile(bgmVolume ~0.24) so funny NPC complaints stay crystal clear. - For
wow_forever_clips, setisolateVocals: trueif streamer Spotify/copyrighted music is audible.
- For
- Pre-Delivery Visual QA (Mandatory):
- After rendering, run
ffprobe+ loudness check (-14 LUFS), then extract 5–8 key frames across every segment and visually inspect them to verify zero black bars, zero cropped top text, and zero caption collisions with baked-in TikTok/Shorts banners (especially during zooms).
- After rendering, run
Handling rejected_feedback
- Never treat
rejected_feedbackas a blind dump or automatic creator/topic ban. - Read every clip's individual
feedback_notecarefully and distinguish between:- Context/timing skips (good clip, just doesn't fit today's montage theme — keep eligible for another theme)
- Format/quality preferences (already over-edited CapCut clip, slow pacing, story spoiler — refine
config/style_guide.mdorconfig/learnings.md) - Explicit source exclusions (only block a creator, tag, or format in
config/scout_queries.jsonwhen specifically requested or a clear multi-clip pattern emerges)
- Always summarize what you learned from
rejected_feedbackbefore applying changes toscout_queries.jsonorstyle_guide.md.
Meme & Pop-Culture Voice Blip Restraint
- Mainstream
<2.5svoice blips over generic SFX: Prioritize recognizable pop-culture voice blips/stingers (<=2.5s, cleanly faded) on real comedic payoffs instead of generic stock booms, pipes, or arcade bells. - Nuance over spam: Iconic game/meme SFX (e.g., Half-Life SFX, Avengers
<2.4shorn sting) work ONLY when peak-synced to exact physical/emotional beats. Overusing a meme or desyncing voice/SFX from on-screen action instantly turns it cringe. - Commercial Music Safety: Never use
>2.5sof clean record-label movie scores — use short<=2.4sstings/parodies or Epidemic Sound tracks instead.
Reference Third-Party Skills
Consult skills/screenwriting-skills, skills/visual-skills, skills/video-editing-skill, and skills/remotion-skills when refining pacing, visual storytelling, or Remotion render components.