Scene Choreography Loader v2.1 - Akki
Get a scene's latest choreography text without hunting through folders
- choreography_text
The AkkiNodes suite generates a lot of intermediate text files, and most of them are versioned with timestamps and counters. If you're running the scene-by-scene production loop, you don't want to hand-edit a filename every iteration - you want a node that says "give me scene 3" and returns the latest choreography file for it. That's Scene Choreography Loader.
It's the read-side counterpart to the suite's save nodes, and it sits in Phase 4 of the pipeline: once the AI Scene Choreographer has produced its per-scene shot prompts, this loader pulls the most recent one back in so you can feed it to a sampler or inspect it. Same pattern as Video Prompt Loader, but keyed on scene number instead of shot name.
How it works
Three inputs:
- project_path - the Project Director's output, or any path relative to ComfyUI's
output/directory. This is a forced input (the little socket), so wire it from the Project Director rather than typing it. - choreography_subfolder (default
DATA) - where choreography files live inside the project folder. - scene_number (default
1, minimum 1) - which scene to fetch.
The node builds output/<project_path>/DATA/, then matches files against a strict naming convention:
CHO_003_2_2026-08-16.txt # scene 003, version 2, dated
The pattern is CHO_ + the scene number zero-padded to three digits + _ + a version number + _ + a date + .txt. It grabs every file matching that scene, sorts them, and returns the newest one's contents as choreography_text. Version, don't overwrite - that's the contract the save side follows, and this loader assumes it.
Installing
Standard for the pack - it's one of ~30 nodes in the suite, so you install the whole thing:
cd ComfyUI/custom_nodes
git clone https://github.com/routhakash/AkkiNodes-LLM-Suite-for-ComfyUI
On Windows, run install.bat as administrator once (it installs llama-cpp-python==0.3.9 and openai into your ComfyUI venv), drop a GGUF model into ComfyUI/models/llms/, and restart. On Linux/macOS, run those pip installs manually - there's no shell installer shipped.
Gotchas
- The filename pattern is non-negotiable. If a choreography file was saved by hand as
CHO_3_final.txtinstead of the padded format, this node won't find it and returnsERROR: No choreography file found for Scene 3 in .... Check the console for the exact path it searched. - It returns error strings, not exceptions. Missing directory or no matching file comes back as a
choreography_textvalue that starts withERROR:. A Text preview node will happily display that, so when your sampler starts choking on prompt text that begins with "ERROR:", look here first. - Scene numbers pad to 3 digits. Scene 3 is
CHO_003, notCHO_3. Thescene_numberinput is an int, but the file lookup always uses%03dformatting. - "Latest" means latest by filename sort, not by file modification time. The convention embeds the date and version in the name, so sort order is reliable - as long as nobody renames files.
This is a niche utility in a niche pack: there's no real community discussion of it, so the source code is the spec. The good news is the behavior is exactly what the code says - deterministic and boring, which is what you want in a loader.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| project_path | STRING | — | |
| choreography_subfolder | STRING | DATA | — |
| scene_number | INT | 1 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| choreography_text | STRING | — |