Communitygithub.com

notasimrazzaq/AI-Video-Editor

Render and export HyperFrames videos: frame rate, 4K, ProRes MOV, PNG sequence, smaller files (--crf), transparent overlays for Premiere/After Effects (PROMPT 14), plus quick fixes when something looks wrong (fonts, cropping, captions overlap, timing, huge files). Use when: the user asks to render, export, change resolution/fps, make it transparent, reduce size, or says something looks off.

What is AI-Video-Editor?

AI-Video-Editor is a Claude Code agent skill that render and export HyperFrames videos: frame rate, 4K, ProRes MOV, PNG sequence, smaller files (--crf), transparent overlays for Premiere/After Effects (PROMPT 14), plus quick fixes when something looks wrong (fonts, cropping, captions overlap, timing, huge files). Use when: the user asks to render, export, change resolution/fps, make it transparent, reduce size, or says something looks off.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/notasimrazzaq/AI-Video-Editor/tree/HEAD/.agents/skills/mg-export

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Render, export, fix

Run from inside projects/<name>/. Final approved files also get copied to exports/ at the workspace root.

Cheat sheet

npx hyperframes render . -o renders/video.mp4                         # default
npx hyperframes render . --fps 60 -o renders/video-60fps.mp4
npx hyperframes render . --resolution 4k -o renders/video-4k.mp4       # aspect must match; portrait → portrait-4k
npx hyperframes render . --resolution 4k --fps 120 -o renders/video-4k-120.mp4
npx hyperframes render . --fps 24 -o renders/video-24fps.mp4
npx hyperframes render . --format mov -o renders/video.mov             # ProRes
npx hyperframes render . --format png-sequence -o renders/frames
npx hyperframes render . --crf 23 -o renders/smaller.mp4               # smaller file
  • When graphics sit on the user's footage, keep the footage fps unless the user asks otherwise.
  • Reference timings on a 4-core i3 laptop with ~12 GB RAM (faster PCs are quicker): 6 s of simple 1080p graphics ≈ 24 s; 3:20 of 1080x1920 footage + Paper Cut ≈ 1.5 h (~1.1 fps: filter: drop-shadow forces screenshot capture, and fast capture only works on macOS anyway). Warn the user before long renders and render once at the end.
  • Speed levers: --workers 3 (each Chrome ~256 MB; default auto picked 1 here — test on a short clip first), --quality draft for review renders, or a shorter clip. If a render runs out of memory or times out, add --low-memory-mode and/or --protocol-timeout 600000.
  • Audio-only changes never need a re-render: node tools/sfx.mjs projects/<n> --mix in.mp4 out.mp4 (video stream copied).
  • Iterate with npx hyperframes snapshot (fast) and render only when frames look right.

Transparent export (PROMPT 14)

MOV/PNG are only transparent if the background is transparent.

  1. Remove the background and any footage so only the card(s) remain; everything around them transparent.
  2. Frosted cards lose their blur with nothing behind them → switch to solid cards for overlays.
  3. Export renders/card.mov (ProRes 4444 with alpha, --format mov) and a PNG sequence in renders/card-png (--format png-sequence).
  4. Snapshot over a checkerboard to verify only the card is visible. (--format webm is another transparent option.)

When something looks wrong

SymptomFix
Wrong font / plain fallbackUse the file in fonts/ via @font-face; never a web font URL. lint flags font_family_without_font_face.
Text cut off on phonesKeep all text inside the middle 80% of the width.
A card covers the captionsMove every card above the bottom third.
A card is a bit early / lateSpeech-to-text can be ~1 s off. Re-time to the exact spoken word (check audio around it).
File is hugeRender again with --crf 23 (grain styles especially).
Silent renderEvery <audio> needs an id; the <video> must not be muted and needs data-has-audio="true".
If the user describes the problem in plain words, translate it to one of these or to a knob in mg-styles.

Individual skills in this repo

This repo contains 4 individual skills — each has its own dedicated page.

Related Skills