Text File Cowboy

If you've ever done per-clip rendering - a folder of 50 clips, each with its own prompt file - you know the pain: a batch loop that has to somehow pick the right text file for the right clip. Text File Cowboy is the TrentNodes answer, and it's part of a family (VideoFolderCowboy, ImageFolderCowboy, RefFolderCowboy) built around filename-keyed batch loading.

The headline feature, straight from the README: wire VideoFolderCowboy's filename output into filename_key, and clip 000.mp4 automatically loads prompts/000.txt. The loop iterates, the right prompt rides along with the right video, and you never index-match anything by hand.

How it works

Give it a directory and it lists the text files there in natural sort order (the same vid1 < vid2 < vid10 ordering the other Cowboys use, so file 10 comes after file 2, not before). Then it picks one file by one of two routes:

• filename_key (optional) - if connected, loads directory/<filename_key><extension>. This is the pairing magic.
• index - a plain index into the sorted list, used only when filename_key isn't connected. index_mode controls overflow: wrap, clamp, or error.

The extension input defaults to .txt - so you could point it at a folder of .md or .json prompt files without a second thought.

Three outputs: text (the file contents - wire this into a CLIP text encode or wherever the prompt goes), filename (bare filename without extension, handy for downstream naming), and total_files (how many text files were found, so the loop knows its length).

The detail that saves you actual time

The README calls out that edits to the file bust the cache automatically via mtime. In ComfyUI's aggressive caching world, most file loaders will happily serve you a stale prompt until you restart or bypass something. Text File Cowboy watches the file's modification time, so you can tweak 023.txt in an editor and re-run - you get the new text without fighting the cache. That one behavior is worth the install by itself if you iterate on prompts a lot.

Where this fits in a real pipeline

The canonical setup, from the README's own framing:

1. VideoFolderCowboy iterates the clips (natural sort, 000.mp4 first)
2. Its filename output feeds TextFileCowboy's filename_key
3. TextFileCowboy loads prompts/000.txt for that clip
4. Text goes to the text encoder, video goes to the video path, frames render per-clip with per-clip prompts

It also pairs with RefFolderCowboy if your per-clip pipeline includes reference images (refs/000/). Together they're the "per-clip render pipeline" the pack is built around.

Gotchas

• If filename_key is connected, index is ignored - wire one or the other, don't fight both.
• wrap overflow silently loops back to the start, which is fine for testing and dangerous for a real batch you thought was length-checked. Use error mode when correctness beats convenience.
• It reads one file at a time, not a whole directory into a list. If you want all files as a batch, that's a different node's job.

Install

One of ~69 nodes in TrentNodes:

# ComfyUI Manager: search "Trent Nodes"

# or:
cd ComfyUI/custom_nodes
git clone https://github.com/TrentHunter82/TrentNodes.git
cd TrentNodes && pip install -r requirements.txt

No extra dependencies or model downloads. It's the kind of node that reads as boring until you've hand-matched 40 clips to 40 prompts once, and then you never go back.