Prompt Preset Selector (Image)

Every ComfyUI preset node has the same blind spot: it hands you a prompt string, and that's it. But half the time a prompt isn't enough on its own - you want a character preset and the pose reference to run through ControlNet, or a style entry and the source image to feed an img2img pass. That's the gap this node fills: Prompt Preset Selector (Image) is the same preset-selector machinery as the rest of this pack, but each YAML entry can carry an image folder along with it, and the node outputs the matching image as a real IMAGE tensor.

One honest caveat before anything else: the README doesn't mention this node at all. It documents four nodes and this is a fifth, registered in the code and discoverable in ComfyUI but effectively undocumented. What I'm about to describe comes from reading the source, so treat this page as the closest thing to docs it has.

How it works

You write your presets as YAML, and link images to keys with a [folder] suffix:

heroes[hero_poses]:
  - superman, red cape, blue suit
  - batman, dark costume, mask

When a line under that key is selected, the node looks in presets/images/hero_poses/ (inside this node's own directory), lists the .png/.jpg/.jpeg/.webp files there, and picks one by image_index. It loads it as an RGB tensor, ready to wire into anything that accepts an IMAGE input, and also hands you the absolute image_path as a string. The [folder] bit is stripped out of the key before the text output, so your prompt comes out clean - heroes: superman, red cape, blue suit, not heroes[hero_poses]: superman.... If a key has no image folder, or the folder is empty or missing, you get a 1x1 black image and "No image" as the path - silently, so check selected_info if your reference frame shows up black.

The inputs and outputs

Everything from the pack's Wildcard node is here - preset_file, absolute_path, keyword, keyword_mode, selection_mode (Manual / Sequential / Sequential (continue) / Random), preset_index, seed, enable_wildcard - plus one extra:

• image_index - picks which image in the folder to use; it wraps around with modulo, so cranking it past the file count just loops back to the start.

Outputs, and this is where it differs from its siblings:

• text - the selected prompt (wire to CLIP Text Encode).
• image - the IMAGE tensor, for img2img, ControlNet, or just a preview.
• image_path - the absolute path of the image on disk, handy if you want to reference it elsewhere.
• preset_list and selected_info - the usual reference and debug strings.

It's filed under its own category, promptPresetSelector, rather than the pack's usual text - mildly annoying when you're hunting for it in the node menu, but you'll only need to find it once.

Install

Same pack, same routine:

cd ComfyUI/custom_nodes
git clone https://github.com/shin131002/ComfyUI-Prompt-Preset-Selector.git
pip install pyyaml   # optional, for YAML files

Then restart ComfyUI. The image loading leans on Pillow, numpy, and torch - all already present in any working ComfyUI install, so there's nothing extra to fetch.

Where people get burned

The black-image fallback is the big one: no folder, empty folder, or a typo in the [folder] name all produce a black 1x1 frame without raising an error, and the workflow happily continues. If you ever get a black reference image, read selected_info - it names the missing folder. Second, images have to live under presets/images/<folder>/ inside this node's directory; it won't go hunting through arbitrary absolute paths like the other inputs do. And given the "no individual support" disclaimer this pack ships with, an undocumented node like this one is the definition of use-at-your-own-risk - but for bundling prompt-plus-reference in one reusable YAML entry, it does something nothing else in the pack does.