Thanks for your interest. Difforum is MIT-licensed and built to be extended.
-
Keep dependencies minimal. Core logic uses numpy and a safe AST evaluator; torch is only used inside nodes that need it. No librosa / numexpr / pandas, so the package keeps installing cleanly on Python 3.12 and the Comfy Registry.
-
Run the tests before opening a PR (no GPU needed):
for t in core audio hybrid models warp color effects look video prompt plot integration; do python tests/test_$t.py done
-
Match the surrounding style (naming, comment density, no em dashes).
-
Regenerate the example workflows if you change a node's inputs:
python examples/_build_examples.pythenpython examples/_validate_workflows.py.
To keep the project's licensing clean and flexible, contributions are accepted under the DCO: you certify that you wrote the patch (or have the right to submit it) and agree it can ship under the project's MIT license. Sign off each commit:
git commit -s -m "your message"which adds a Signed-off-by: Your Name <you@example.com> line.
Good places to extend: new schedule functions, camera presets, effects (the
core/effects.py and core/symmetry.py style), reactive audio modes, model
catalog recipes, and the realtime path (NDI sink, MIDI/OSC control). See
DESIGN.md and REALTIME.md.