Skip to content

insodimension/dopamine-debt

v0.5.0

Makes deadpan, meme-led comedy explainer films about your channel's subject, and checks every stage against house rules before you spend on voice and music.

GitHub

Dopamine Debt

What it is

The Dopamine Debt kit: a framework, a Filmmaker General Agent and a film kit for deadpan, meme-led comedy explainer films about your own channel's subject. It takes a film from audience research through spine, joke map, script, voice, picture and cut to a render, and holds every stage to the house gate (inso dd gate) before the next one starts. The Video space requires it, so installing Video installs it from this store.

What it contains

  • General Agent filmmaker: makes a film end to end in a Video production's room. On by default; the Video space lists it in its agent picker.
  • Skills filmmaker-core (the order of work, loaded by the Filmmaker), dopamine-debt-film (the framework, research to approved render, and turning review notes into rules) and kit-film (building the picture with the film kit).
  • CLI inso dd:
    • inso dd setup <brand folder> writes the brand's productions/house-gates.json, whose gate runs inso dd gate, so an update of the kit never breaks it. An entry written by an earlier kit (one that ran gate/launch.ts) is replaced, and setup says so.
    • inso dd gate <script|cut> <production> <report> runs the house gate.
    • inso dd new-film <production> writes a film's build files; they find the kit each time they run.
    • inso dd import <earlier kit folder> <brand folder> carries a music ledger, review marks and personas from an earlier copy of the kit into the brand's folder.
    • inso dd root prints the installed kit's folder.
  • The engine and data: the Python engine (dd/, framework/ and its film kit), a starting channel persona for an AI-news audience, taste.json (the house rules), tunables.json (the gate's constants) and an empty music.json.

Who can use it

Anyone with Dimension 0.12.0 or newer. It installs from the Community store like any pack, and the Video space requires it, so installing Video installs it. Its Filmmaker appears only where a space lists it (Video does). It needs:

  • Python 3.10 or newer on PATH, or DD_PYTHON set to one, with the packages in requirements.txt (python -m pip install -r <folder>/requirements.txt, where inso dd root prints the folder). inso dd gate exits 2 and names what is missing.
  • ffmpeg, a headless Chrome or Edge, npx hyperframes, and the Motion pack's setup (it installs GSAP) for the picture and the render.
  • The inso command, which Dimension's installer puts on PATH. It hosts the gate's judge model and runs the gate; there is nothing else to install.

Limits and risks

  • What it reads and writes: the production folder it is given; the brand's productions/ folder (house-gates.json) and its productions/dopamine-debt/ folder, where everything you make for the kit lives (films, research, your personas, music.json, calibration/marks.jsonl); the shared media library; and a speech-recognition cache in <engine home>/productions/dopamine-debt/. It never writes into its own folder, which an update replaces.
  • What it sends over the network: the gate's script and transcript text to the model your judge role names; npx -y hyperframes@0.8.217 downloads HyperFrames from npm on the first render; faster-whisper downloads its small.en model from Hugging Face on first use; media fetches go to YouTube through yt-dlp (only for http(s) addresses you or the Filmmaker chose). Voice and music generation go through Dimension's media generation tools, priced and approved before each spend.
  • The gate judges with a model, so a gate run spends model tokens; its report names the model that answered.
  • It does not install its Python packages, ffmpeg or a browser for you.
  • The house rules come from one channel's films and one reviewer's taste; your brand gets the same bar until your own notes and marks move it.
  • The starting channel persona is a proxy for an AI-news audience, built from public comment counts and summarised in our own words. Build your own channel persona once your films have viewers.
  • It ships no calibration films or research corpus. The skills' Calibrate step reads your own films, kept in your brand's productions/dopamine-debt/calibration/.

Build and test

No build step: the host loads the CLI's TypeScript as it is, and it imports only modules the host provides. The engine is Python. Its unit tests live in tests/: python -m pytest tests in this folder (bun run test), with the packages in requirements.txt and pytest. bun run check:types typechecks the CLI from a Dimension monorepo checkout. The calibration films the gate was fitted on are not in this repository.