| name | prismtek-youtube-buddy-lipsync |
| description | Produce a Prismtek/Buddy talking-host overlay for narrated videos using pixel-art visemes and FFmpeg audio RMS animation. |
| version | 1.1.0 |
| author | Prismtek |
| license | MIT |
| platforms | ["macos","linux"] |
| metadata | {"hermes":{"tags":["youtube","video","ffmpeg","imagemagick","pixel-art","vtuber","buddy","prismtek"],"related_skills":["youtube-content","media-rendering","buddy-appearance"],"safety":{"stores_credentials":false,"uploads_to_youtube":false,"commits_media_assets":false}}} |
Prismtek YouTube Buddy Lip-Sync
Overview
This skill turns a clean Buddy host PNG into a lightweight audio-reactive talking-host overlay for Prismtek/Buddy videos.
It is intentionally boring in the best way:
- no ML lip-sync dependency;
- no face-swap, cloning, impersonation, or deepfake pipeline;
- no YouTube upload automation;
- no OAuth tokens, cookies, videos, receipts, or private assets committed;
- ImageMagick for deterministic pixel-art mouth/viseme generation;
- FFmpeg/ffprobe for audio RMS analysis and video compositing;
- JSON receipts for repeatable render review.
When to Use
Use this when:
- a Prismtek/Buddy video has narration and Buddy should visibly speak;
- a static Buddy overlay feels too lifeless;
- you need proof clips before full rendering;
- you need a reproducible pixel-art mouth workflow that avoids visual regressions.
Do not use this for deceptive impersonation, face-swapping a real person, or uploading to accounts without explicit operator approval.
Required Tools
command -v magick
command -v ffmpeg
command -v ffprobe
python3 --version
The scripts intentionally avoid PIL/Pillow so the workflow works in small runtimes.
Recommended Asset Layout
Keep production assets outside git:
~/.hermes/assets/prismtek-youtube/avatar/buddy-host-main-clean.png
~/.hermes/outputs/prismtek-youtube/
The checked-in skill stores the reusable workflow only. Generated PNGs, rendered MP4s, receipts, and source videos should stay in local output folders or private artifact storage.
Generate Buddy Visemes
python3 scripts/generate_buddy_v12_visemes.py \
~/.hermes/assets/prismtek-youtube/avatar/buddy-host-main-clean.png \
--out-dir ~/.hermes/assets/prismtek-youtube/avatar
Outputs:
buddy-host-main-v12-no-mouth-base.png
buddy-host-main-v12-mouth-closed.png
buddy-host-main-v12-mouth-small-open.png
buddy-host-main-v12-mouth-wide-open.png
buddy-host-main-v12-mouth-round-o.png
buddy-host-main-v12-mouth-soft.png
Render a Proof Clip
python3 scripts/prismtek_youtube_avatar_lipsync.py \
/path/to/base-video.mp4 \
--closed ~/.hermes/assets/prismtek-youtube/avatar/buddy-host-main-v12-mouth-closed.png \
--open ~/.hermes/assets/prismtek-youtube/avatar/buddy-host-main-v12-mouth-small-open.png \
--output /tmp/buddy-lipsync-proof.mp4 \
--limit-seconds 20
For full render, omit --limit-seconds.
v12 Mouth Style Rules
Accepted style:
- small/cute pixel-art mouth;
- slightly larger than the first tiny-mouth attempt;
- one visible mouth only;
- no darker mint rectangle around the mouth;
- no original-smile leftovers creating a second mouth;
- full Buddy crop remains intact.
Rejected styles:
- oversized oval or Pac-Man mouth;
- darker mint erase patch;
- two-mouth artifacts from leftover smile pixels;
- cropped arms, source-sheet borders, or sparkles.
Verification Checklist
Before publishing, render a proof clip and inspect frames around speech and silence:
ffmpeg -y -v error -ss 00:00:05 -i /tmp/buddy-lipsync-proof.mp4 -frames:v 1 /tmp/buddy-open.png
ffmpeg -y -v error -ss 00:00:10 -i /tmp/buddy-lipsync-proof.mp4 -frames:v 1 /tmp/buddy-closed.png
ffprobe -v error -show_entries format=duration -show_entries stream=codec_type,codec_name,width,height -of compact=p=0:nk=1 /tmp/buddy-lipsync-proof.mp4
Visual QA:
- Buddy has one mouth only;
- mouth opens/closes with narration energy;
- no darker face patch appears around the mouth;
- Buddy remains fully visible in the chosen overlay position;
- output duration and audio stream are preserved;
- receipt JSON exists next to the output.
Repository Ownership
buddy-agent: runtime-facing skill copy.buddy-brain: operator/source-of-truth skill copy.omni-buddy: local device/voice host skill copy.prismtek-apps: app-facing documentation and future UI contract reference.knowledge-vault: durable skill reference and operating notes.hermes-agent fork/upstream PR: optional Hermes-compatible media skill.
Safety Notes
This skill does not perform account login, upload to YouTube, scrape private sessions, or commit private media. Keep secrets in environment variables or a secret manager, and keep generated media out of public git history.
