ugc-enhance
Skill4 starsChanged 5 months agoMITView on GitHub
Step 6 (optional, default ON) of the ugc-ad pipeline. Post-produces a finished UGC ad into a leveled-up cut — adds a ducked ElevenLabs music bed, word-level karaoke subtitles synced to the native VO, and flash + zoom-punch transitions on the hard cuts — via a HyperFrames composition. Use standalone to enhance any talking-head/UGC mp4, or as the final step of a UGC ad. Brand-agnostic. Validated 2026-06-22.
- Reads credentials
- Sends data out
In this repository7 other files it gives its agents.
- ugc-adSkill
- ugc-base-characterSkill
- ugc-briefSkill
- ugc-multicut-scriptSkill
- ugc-product-profileSkill
- ugc-storyboard-sheetSkill
- ugc-videoSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Music bed: If MUSIC given, copy it to $DEST/enhance/music.mp3.
- Very specific: line 30
Verify it is real audio (file music.mp3 -> MPEG/ID3, not a JSON error).
- Distinctive: line 64
No em-dashes / en-dashes in any on-screen text.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
ugc-enhance
Turns the bare Seedance master into a music-bedded, subtitled, cut-punched social cut. Local render (no Seedance credits); the only paid call is the ElevenLabs music gen.
Inputs
MASTER— path to the finished base ad mp4 (the native-VO Seedance output).DEST— the project folder (writes<DEST>/enhance/working dir +<DEST>/<slug>-enhanced.mp4).SLUG.- optional
BRAND_ACCENT(hex; default#2BA8E0— pull from the product label / brand cues). - optional
MUSIC— path to a provided track (skip generation); else ElevenLabs generates one. - optional
MUSIC_PROMPT,MUSIC_VOL(default0.16).
Steps
- Working dir.
mkdir -p "$DEST/enhance"; copyMASTER->$DEST/enhance/video.mp4. Get duration:ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 video.mp4. - Music bed. If
MUSICgiven, copy it to$DEST/enhance/music.mp3. Else generate with ElevenLabs (key in.env.localasELEVENLABS_API_KEY; read it without echoing):
Verify it is real audio (KEY=$(grep -E '^ELEVENLABS_API_KEY=' .env.local | cut -d= -f2- | tr -d '"'\''') curl -s -X POST "https://api.elevenlabs.io/v1/music" \ -H "xi-api-key: $KEY" -H "Content-Type: application/json" \ -d '{"prompt":"<vibe matched to the ad: e.g. clean optimistic corporate tech, light electronic, no vocals, trustworthy product-ad bed>","music_length_ms":<round(duration*1000)+1000>}' \ -o "$DEST/enhance/music.mp3"file music.mp3-> MPEG/ID3, not a JSON error). No vocals (they fight the VO). - Transcribe the VO for caption timing (English talking-head): in
$DEST/enhance/,npx hyperframes transcribe video.mp4 --model small.en. Readtranscript.json; run the transcript quality check (strip music/garble tokens). Word-level timing is required for karaoke. - Detect hard cuts:
These timestamps becomeffmpeg -i "$DEST/enhance/video.mp4" -filter:v "select='gt(scene,0.3)',showinfo" -f null - 2>&1 | grep -oE "pts_time:[0-9.]+"window.__CUTS(flash + zoom-punch land here). - Author
captions-data.jsin$DEST/enhance/from the transcript. Read assets/captions-data.example.js for the exact shape and the grouping/accent/brand rules. Key rules:- Group by sentence end, a >0.4s pause, or max ~5 words; one strong one-word sentence can stand alone.
accent: trueon brand names, figures, and key claim words;num: trueon figures.- Brand display rule: the VO is phonetically respelled (e.g. "Noo Standard Labz") but the caption MUST show the real brand spelling (e.g. "Nustandard Labs"). Map spoken words to correct display text, keeping the spoken timing.
- Set
window.__ACCENTtoBRAND_ACCENTandwindow.__CUTSto the detected cuts.
- Assemble the composition. Copy assets/index.html.template ->
$DEST/enhance/index.htmland assets/DESIGN.md ->$DEST/enhance/DESIGN.md. Replace{{DURATION}}(every occurrence) with the video duration and{{MUSIC_VOL}}withMUSIC_VOL. - Render. In
$DEST/enhance/:
(Contrast audit is skipped — captions ride over live video frames.)npx hyperframes validate --no-contrast npx hyperframes render . -o "../<slug>-enhanced.mp4" --quality high - CHECKPOINT (human). Open
$DEST/<slug>-enhanced.mp4. Only a human can verify the audio mix (music ducked under VO) and the captions (timing + the brand spelling on screen). AdjustMUSIC_VOL, accent, grouping, or transitions and re-render (free) until right.
Output
$DEST/<slug>-enhanced.mp4 (the leveled-up cut) + the re-renderable $DEST/enhance/ recipe (index.html, DESIGN.md, captions-data.js, music.mp3, transcript.json).
Constraints
- Music is a bed: ducked under the native VO (
MUSIC_VOL~0.16), never vocals. - Captions ride the lower third; never cover the creator's face. One group visible at a time; every group hard-killed at its end.
- Brand name on screen uses the REAL spelling, not the phonetic VO spelling.
- No em-dashes / en-dashes in any on-screen text.
- Local render only — no Seedance credits. The base master is left untouched (enhanced cut is a separate file).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

