Dawnwood / Open source
OPEN SOURCE · APACHE-2.0

The tools behind our shows. Free to use.

Dawnwood is an AI drama studio. Everything here runs in our own production line every day. Take it into yours: commercial use is fine, and we only ask for a credit.

PROJECTS

Ready to use

dramapackPYTHON · CLI · MCPAPACHE-2.0
From finished episodes to a delivery pack a distributor or platform accepts. It covers the parts of delivery that every short-drama studio ends up redoing by hand:
  • Mastering: subtitles burned inside the platform safe area (text box at most 70% of the frame width, clear of the player bar), loudness set to -14 LUFS with a constant gain rather than dynamic loudnorm, and an optional procedural music bed that ducks under dialogue.
  • Delivery pack: numbered episodes with same-name SRT, a clean master with no subtitles, 3:4 covers, a timeline and EDL, chain-of-title and AI-disclosure documents, and a manifest with measured specs.
  • Acceptance audit: red, yellow or green against short-drama market benchmarks before anything ships.
  • Rights check: flags undeclared or major-label music and fonts that aren't licensed for commercial use.
  • For agents: an MCP server, --json on every command, plus AGENTS.md, llms.txt and a Claude Code skill.
pip install dramapack
dramapack package ./episodes ./delivery
# or hand it to your agent:
claude mcp add dramapack -- dramapack mcp
demusicPYTHON · CLI · MCPAPACHE-2.0NEW · NOV 2026
Removes the background music that AI video models (Seedance, Kling, Veo, Hailuo, Wan) bake into generated clips, and keeps the dialogue, sound effects and ambience.
  • Finds where the music is: stem separation (Demucs, local or on your own fal key), a loudness gate, then an AudioSet music classifier that has to hear music for two seconds in a row. Engine hum, gunfire and room tone don't count.
  • Only touches those seconds: inside the detected spans the audio becomes dialogue plus impacts, crossfaded; everywhere else the original audio stays and the picture is stream-copied.
  • Conservative by design: 2 false positives on 164 clean production clips. Lower the threshold or force it when you know a clip has music.
  • For pipelines and agents: demusic check exits 1 when a clip has music, --json output, MCP server, AGENTS.md and a Claude Code skill.
pip install "demusic[demucs]"
demusic strip renders/*.mp4 --out-dir clean/
# or hand it to your agent:
claude mcp add demusic -- demusic mcp
NEXT

Being extracted from our pipeline

LOCALIZATION
Multi-language dubbing
One series, many language versions: dubbing, lip-sync handoff, subtitle reflow and remastering, with the failure modes we've already hit handled.
QUALITY GATES
Render QC
Automated checks that catch bad takes before a person has to watch them: visual verification and headshot framing.
QUESTIONS

The short version

Is it really free?
Yes. Every project here is Apache-2.0: commercial use, your own production line, modifications. We only ask that you keep the generator credit it writes into delivery manifests, and mention dramapack by Dawnwood when you describe your pipeline.
Can my AI agent use it?
Yes. dramapack ships an MCP server (dramapack mcp), every command has --json output, and the repo includes AGENTS.md, llms.txt and a Claude Code skill.
Why would a drama studio give its tools away?
We make AI drama series, and these tools run on every episode we ship. Delivery, loudness and subtitle safe areas are problems every studio solves again. Sharing our fixes costs us nothing, and it gives distributors something concrete to check when we co-produce or license with them. Want to work together? Work with us.