Advanced Result Browser

Advanced Result Browser is the batch version of the pack's basic Result Browser. Where the plain node hands you one image at a time for manual scrolling, this one outputs a whole stack of the top matches at once - plus their filenames and scores - in a single run. If you want to dump the 20 best matches into a grid viewer, feed them into an upscaler, or review a contact sheet, this is the node.

It's the same search pipeline (Image Searcher → this node) with more knobs, and it adds two things the basic browser lacks: a similarity threshold cutoff and a batch of images instead of one.

The inputs that matter

• results - the SRESULTS output from Image Searcher.
• image_count - how many images to output (default 1, so bump it - that's the whole point of this node).
• offset_index - where in the ranked list to start, same paging idea as the basic browser's image_index.
• match - first (most similar at the front) or last (least similar at the front).
• threshold - a similarity floor. Images scoring below it stop the batch (default -1, which effectively disables the cutoff since cosine scores rarely go negative). Set it to e.g. 0.5 to only collect confident matches.
• batch_frame_image (optional) - an IMAGE input used as a resolution reference. The batch is resized to match this frame's dimensions, and the reference image itself is stripped from the output afterward.

The outputs

Three, same names as the basic browser but with different shapes:

• IMAGE - an actual image batch of all collected results, ready for a batch-aware viewer or any node that consumes a stack of images.
• FILENAME - the list of stored filenames (as a string list, one per result).
• SCORE - the list of similarity scores (a float list), matching the image order.

The mechanism is worth knowing: it sorts the distances, walks the list from offset_index, and stops when it hits image_count images or a score below threshold, whichever comes first. Images are converted to RGB as they're loaded (HEIC and JPEG weirdness handled by the pack's dependencies).

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/MoonMoon82/ClipVision_Tools
cd ClipVision_Tools
python -m pip install -r requirements.txt

Or ComfyUI Manager → "ClipVision_Tools". Deps: orjson, pillow-heif.

Gotchas

• Threshold is a stop, not a filter. The node breaks out of its loop on the first below-threshold score; since scores are sorted, that's equivalent to filtering, but it means a badly ranked early entry can truncate your batch. The representative-score quirks in FolderScores don't apply here - this one's straightforward.
• List outputs vs. single values: don't wire FILENAME or SCORE into nodes expecting one string/float without handling the list-ness first.
• Resolution reference: if you pass batch_frame_image, every result gets resized to it. Useful for consistent grids; confusing if you forgot you passed it and wonder why everything came out 512×512.

For review-and-pick work - "here are the 20 closest, choose" - this is the node you want over the basic browser. It's the same underlying search, just less clicky.