Load Text Files from ZIP

Load Text Files from ZIP reads every text file inside a zip archive and hands you two views of them at once: the whole pile as lists, and a per-file stream that runs everything downstream once per file. That second part is the killer feature. You know how training data ships as folders of images with a same-named .txt caption beside each one? This is how you feed that exact arrangement to a sampler without unpacking anything - one file becomes one generation, caption and all.

It's a WAS Node Suite node (find it under WAS Suite/Archive), part of the pack's files-and-archives family that treats a zip as something to read on the wire instead of a thing you excavate first.

How it works

Point the file menu at an archive (it lists every .zip in ComfyUI's input, output, and temp folders, tagged with where it lives) and set pattern. * takes every text file at any depth; *.txt takes every .txt in any folder; cat_*.txt takes only those named that way. A / anchors the pattern at the archive root, so captions/**/*.txt reads one folder and everything under it. Case is ignored. Like its image sibling, it can take a ZIP socket from Open ZIP, which overrides the menu and opens the archive once however many nodes read it.

Each file comes out in two shapes because the two shapes serve two different graphs:

• texts (an ARRAY) - every file's text on one wire, index-aligned with names. Grab one with Text List Get when you want to fetch a specific caption.
• text (a STRING list) - the same texts, but wired into something like a sampler's prompt, ComfyUI runs the whole graph once per entry. That's the batch loop. Its partner name carries the matching filename each run, so you can feed it into Image Save's filename prefix and every render comes out named after the caption that made it.

That "runs downstream once per file" behaviour is the thing most people don't expect from a loader, and it's what makes caption-driven batch prompting possible without a single loop node.

The other outputs

count tells you how many files were read, and skipped how many entries didn't make it - wrong kind of file, an unsafe name, a symlink, encrypted, damaged, or not UTF-8 text. Every rejection is named in the log, so a silent gap is easy to trace. If the archive is missing, isn't readable, or holds nothing your pattern picks, the prompt stops and says which.

Installing it

Same routine as every WAS node: ComfyUI Manager → "WAS Node Suite v3", or

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/was-node-suite-comfyui.git

restart, done. Requires ComfyUI 0.14.0+ and Python 3.10+, and the pack installs no dependencies of its own.

Where people get burned

The pattern is the trap. * matches at any depth, so it quietly grabs every .txt (and .json, and anything text-shaped) the archive holds - if your zip mixes captions with readme files, be specific. And remember the text output is a list wire: if you wire it into a node that doesn't iterate lists, you'll only see the first (or last) file's contents. If you want one caption per run instead of the whole archive, use texts/names with a picker rather than the list output. Oh, and files a zip was encrypted with won't read - that's a "skipped" entry, not a hang.