- a ComfyUI that runs (developed against 0.38 and later);
- git.
1. Clone, and link the node pack into ComfyUI
Windows, in an ordinary Command Prompt (mklink /J needs no administrator rights):
git pull updates what ComfyUI loads.
2. Start ComfyUI
seriesmith shows up in ComfyUI:- seriesmith Shots and seriesmith Refs in the sidebar;
- seriesmith Timeline in the bottom panel.
Or: the editor on a page of its own
With ComfyUI running, double-clickrun_seriesmith.bat in the repo (Linux/macOS:
./run_seriesmith.sh).
- The first time, it sets itself up in the repo’s
.venv. That takes a minute or two and needs Python 3.10 or later. - Then it opens http://127.0.0.1:8190 in your browser.
Which models it uses
When seriesmith starts and sees ComfyUI, it checks which models have all their files installed. A series that doesn’t name a model renders on the first one it finds:- New projects: new series and Simple-mode projects write the choice into their series config, so they keep rendering on it.
- Your own pick: choose another in Settings (Video model, Picture model).
- Paid models: hosted Comfy Router models (Seedance, Kling, Veo, …) are never picked by themselves. Add a Comfy API key in Settings to use them (and Comfy Cloud). Saving the key also installs what those need (Comfy’s SDK) in the background, in a few seconds.
seriesmith targets prints the same list from the command line. The smallest
complete set is the default pair: MiniMax H3 Ref2VA for video and Krea 2 for pictures
(INSTALL.md, The minimum set).
What works now, and what’s extra
With just this, you can do all of the following on the models you have:- write and build episodes;
- render shots;
- make reference pictures;
- cut the episode;
- export it.
- MiniMax H3 at full quality (its sparse-attention node);
- generated voice samples;
- upscaling;
- post-processing (enhancement, motion blur);
- matching recorded dialogue;
- 360s of a place, and turning a character.