🌻 Text Saver

Writes a string to a text file. That's the job - but it's the piece that lets a ComfyUI graph save text the way it saves images, which turns out to matter a lot once you're captioning, logging, or building datasets. Give it a string and a folder and it drops a numbered .txt next to your other outputs.

Where it fits

ComfyUI is built to output images; saving text is an afterthought it doesn't do natively. But plenty of workflows produce text worth keeping - a caption from a vision model, a generated description, a log line, an assembled prompt. The obvious pairing is captioning for training: run an image captioner, then Text Saver writes each caption to a .txt, and if you match its naming to your Image Saver you get image/caption pairs (img_0001.png + img_0001.txt), which is exactly the layout LoRA trainers expect.

How it works

It writes the incoming text to the directory you name, building the filename from filename_prefix, filename_delimiter, and filename_number_padding with automatic numbering. The optional file_extension lets you write something other than .txt (say .json), and encoding defaults to utf-8. Per the pack's README, the directory and prefix also support time-based tokens, so you can bucket outputs by date if you set them up that way.

The inputs that matter

• text - the string to write. Wire it from whatever produced the text.
• directory - where the file goes. Defaults to ./ComfyUI/output.
• filename_prefix / filename_delimiter / filename_number_padding - the naming scheme (e.g. ComfyUI_0001). Match these to your Image Saver for paired files.
• file_extension (optional) - .txt by default; change it for other formats.
• encoding (optional) - utf-8 unless you have a reason otherwise.

The output is File_Path, the full path it wrote - handy to pass into a zip or cleanup node downstream.

Common issues

If you're building image/caption pairs, the naming scheme has to match your image saver exactly - same prefix, delimiter, and padding - or the .txt and .png won't line up by name and your dataset loader won't pair them. Keep encoding on utf-8 unless you're dealing with a tool that demands something else; mismatched encoding is how you get mojibake in captions, especially with non-Latin text. And directory is a path on the ComfyUI host, so on a cloud instance the file lands server-side - grab it via File_Path (e.g. into a Zip Save) rather than expecting it on your own machine.

Installing ComfyUI-Addoor

ComfyUI-Addoor (葵花宝典) is a utility pack from developer Eagle-CN. ComfyUI Manager: Install Custom Nodes → search "ComfyUI-Addoor" → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Eagle-CN/ComfyUI-Addoor.git
cd ComfyUI-Addoor
pip install -r requirements.txt

Restart and find it under 🌻 Addoor / Utilities. Pair it with an image captioner and Image Saver to build training datasets.