Y7 Image Batch Path

Most image loaders in ComfyUI hand you tensors and then drop the file on the floor. If you're captioning a dataset, you need the file path as much as the pixels - otherwise you can't write cat.txt next to cat.jpg. Y7 Image Batch Path is the loader that doesn't drop it: it loads a whole directory of images and hands you both the image tensors and a matching list of their full file paths.

It's the front half of a three-node captioning chain in the Y7Nodes pack - Image Batch Path → a VLM captioner (the pack's own Y7 JoyCaption works great) → Y7 Caption Saver. The path list is what tells Caption Saver exactly where to write each .txt file, which is what makes the whole pipeline hands-off.

How it works

Point it at a directory with image_dir (a multiline string - paste a folder path) and it walks the contents, loading jpg, jpeg, png, and webp files. Each image is EXIF-transposed (so phone photos with orientation tags come out right side up) and converted to an RGB float32 tensor - the standard shape ComfyUI nodes expect.

The output is two parallel lists, matched one-to-one:

• IMAGE - the image tensors, one per file.
• IMAGE_PATH - the full file path for each image.

That paired output is the whole point. Wire IMAGE into JoyCaption (or any VLM) for captioning, wire IMAGE_PATH into Caption Saver so the captions land next to their source files. Because both are lists, the batch flows through ComfyUI as a whole, not one-image-at-a-time.

The settings that matter

• batch_size - how many images to load. 0 = all of them, which is the default and usually what you want.
• start_from - 1-based index of the first image to load. This is the resume feature: if you've already captioned the first 50 files and the pipeline died, set start_from to 51 and pick up where you left off without redoing work.
• sort_method - sequential (alphabetical), reverse, or random. One gotcha: random re-evaluates on every run, so if you use it and re-run, you'll get a different order each time. Fine for sampling, wrong for anything where order must be stable.

Pairing it with the rest of the pack

The three-node chain is the intended use:

1. Image Batch Path loads the folder and emits IMAGE + IMAGE_PATH.
2. Y7 JoyCaption takes IMAGE and generates a caption for each.
3. Y7 Caption Saver takes the caption STRING + IMAGE_PATH and writes cat.txt next to cat.jpg.

If you're building a training dataset, this is the loop that does it in one pass. And given the KB's note that natural-language captions (JoyCaption-style) are the standard for LLM-encoder models while Danbooru taggers suit the anime lineage, this node is agnostic - it feeds whatever captioner you prefer.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/yushan777/ComfyUI-Y7Nodes
cd ComfyUI-Y7Nodes
pip install -r requirements.txt

Restart ComfyUI, or use ComfyUI Manager → search "Y7Nodes".

Gotchas

The two things people hit: forgetting that batch_size 0 means "everything" (if the folder is huge, the first run loads it all into memory at once - check your folder size), and expecting a stable order from random. Otherwise it's a solid, boring utility that does exactly one job and does it right.