automanim

chat to make manim vids. preview. save. that’s it.

demo: assets/demo.mp4

why?

writing manim code is hard. llms are good at it. i wanted a tool where i can just say "make a red circle" and it happens. and then "move it right". preserving context is key.

prereqs

• python 3.9 (manim hates 3.10+ sometimes)
• node.js & npm
• ollama (running locally or remote)
• system deps for manim (ffmpeg, pango, libcairo2)

installation

1. clone repo.

git clone https://github.com/niyarrbarman/automanim
cd automanim

2. setup. run the helper script. it handles venv (pip, conda, or uv) and installs npm dependencies.

chmod +x setup.sh run.sh
./setup.sh

running

start everything. launches backend (8000) and frontend (3000).

./run.sh

then open http://localhost:3000.

features

chat & iterate

• context aware: knows what you built last. "make it blue" works.
• smart handling: sanitizes code. uses correct latex syntax (MathTex vs Tex).

scene history

• tree structure: don't like the new change? go back to parent node and try again.
• branching: explore different variations from the same point.
• actions: edit prompts. delete mistakes. retry generations.

workspace

• multi-tab: work on 'calculus.py' and 'algebra.py' at the same time.
• auto-save: localstorage keeps your work safe.
• code editor: syntax highlighting. auto-wrap. copy/download code.

models

• pluggable: built for ollama.
• auto-pull: select a model in dropdown. if missing, it pulls automatically.
• defaults: supported models include:
  • gpt-oss:120b-cloud (default)
  • gemini-3-flash-preview:cloud
  • qwen3-coder:480b-cloud
  • qwen3-next:80b-cloud
  • gemma3:27b-cloud
  • glm-4.7:cloud
  • ministral-3:14b-cloud

configuration

frontend uses .env.local (created by setup.sh). backend uses standard env vars.

backend options:

• LLM_PROVIDER: ollama (default)
• OLLAMA_HOST: default http://localhost:11434
• OLLAMA_MODEL: default gpt-oss:120b-cloud (change via ui or env)

troubleshooting

• manim crash? usually missing system libs. install libcairo2-dev libpango1.0-dev ffmpeg.
• video not playing? browser cache. we use timestamps to bust it, but try hard refresh (ctrl+shift+r).
• model timeout? first run might be slow if pulling massive models. check terminal output.

license

mit. do whatever.

todo / known bugs

• preview sync: sometimes video preview doesn't update immediately when switching between projects. (hard refresh fixes it).
• local models: currently UI only lists cloud models. need to fetch local ollama list.