A small local web interface for generating and curating short audio loops with
Magenta RealTime 2 (mrt) on Apple Silicon. Type a prompt, pull four faders
(Outputs / Length / Randomness / Complexity), generate a batch of variations,
audition them in-browser, keep the good ones, bin the rest. Unofficial
third-party front-end over Google's mrt CLI — does not bundle or modify
Magenta itself.
loop_lab.py— the entire app (local web server + generation logic)README.md— setup, requirements (Apple Silicon, Python 3.12, magenta-rt), usagerequirements.txt— Python dependenciesdocs/—screenshot.png(UI reference),Loop Lab Recording.skbundle/.mov(demo screen recording),Loop Lab - final.mp4(recording with title + end cards),loop-lab-demo.mp4(4.6 MB web/README version); promo-card sources live in~/ScreenKiteWorkspace/loop-lab-cards/
- Unofficial / not affiliated with Google. Keep that framing in any user-facing copy. Don't imply endorsement.
- Generated audio (
*.wav),outputs/, andmanifest.jsonare gitignored — they are artifacts, not source. Don't commit them. - Brand: dark UI, magenta accent
#EC1E8C, bold "LOOP LAB" wordmark with the second "O" in magenta, mono labels. GitHub: github.com/mgoldhack/loop-lab
- Run the app: see
README.md(needs magenta-rt installed + models downloaded in a Python 3.12 venv;mrt models download). - Loops are generated per-prompt at varied temperature so a batch actually differs (the model uses a fixed seed).
- Demo video editing is done in ScreenKite (see SESSION_LOG for the promo-card
work and the trailing-overlay limitation). Captions were added outside
ScreenKite via headless-Chrome PNGs +
ffmpeg overlay(this ffmpeg has no drawtext/libass) — see memoryvideo-captioning-ffmpeg.