📂 Project Plan Loader
How the second flow reads the plan the first flow's LLM already wrote
- agent_json
- scene_count
- plan_path
If you've looked at the pack's DynamicSceneDirector, you may have wondered about the split-workflow pattern it implies: Flow 1 runs an LLM to turn a script into a JSON scene plan, and Flow 2 - the heavy video render - shouldn't have to re-run that LLM on every chunk. ProjectPlanLoader is the reading end of that bargain. It loads the plan JSON Flow 1 saved to disk and hands it to Flow 2's Director, no model, no API call, no repetition.
The tooltip says it all: connect the audio_filename input to the current_file_name output of the pack's Audio Batch Selector. The filename isn't a label here - it's the key that locates the plan file.
How it works
It rebuilds the exact same path the DynamicSceneDirector uses when saving: take audio_filename, strip it to alphanumerics plus spaces, underscore, and dash (so my_audio.mp3 becomes my_audio), and look for <output_dir>/<name>_plan.json. The output directory is ComfyUI's standard output folder, so the file lives somewhere you can actually inspect. It reads the raw JSON, parses it, counts the scenes array, and returns all three to you.
Two design details worth knowing. It's IS_CHANGED-always-true, meaning it deliberately re-reads the file from disk every time - you can edit the plan in a text editor mid-render and the next chunk picks up your change. And if the plan file is missing, it raises a hard error instead of silently returning nothing. That's the intended fail-fast behavior; the error message even tells you to run Flow 1 first.
Inputs and outputs
audio_filename(STRING) - from the Audio Batch Selector'scurrent_file_name, or type it to match whatever Flow 1 used.
Outputs:
agent_json(STRING) - the raw plan JSON, wired into the Director'sagent_jsoninput.scene_count(INT) - how many scenes the plan has.plan_path(STRING) - full path to the file, useful for debugging or chaining.
Installing it
It's part of comfyui-sequential-batcher:
cd ComfyUI/custom_nodes
git clone https://github.com/Meisoftcoltd/comfyui-sequential-batcher
or ComfyUI Manager, search "comfyui-sequential-batcher", restart. No extra dependencies - this node is pure filesystem + JSON.
Common issues
The one error you'll actually hit is the missing-plan failure: No se encontró el plan de proyecto. It means Flow 1 never ran (or saved under a different audio filename). Make sure both flows use the same audio_filename, and remember the sanitization - scene_01.wav and scene 01.wav resolve to different keys than you might expect. There's no LLM here to fall back on, so the plan file genuinely has to exist.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| audio_filename | STRING | Conectar al 'current_file_name' del Audio Batch Selector |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| agent_json | STRING | — |
| scene_count | INT | — |
| plan_path | STRING | — |