Sample Tags

Paste in a pool of tags, tell it how many to grab, and Sample Tags hands you back a random subset - no duplicates, count controlled by a range rather than a fixed number. It's wildcard-style randomization, but as a visible node with actual controls instead of {a|b|c} syntax buried inside your prompt text.

How it works

tags is a multiline block of tags, split by whichever tags_delimiter you pick - newline or comma. The node then samples somewhere between min_k and max_k of them, without repeats, and joins the picks back into a string. seed controls the draw: same seed against the same tag list gives you the same subset every time, so it's reproducible if you need it to be, or wire a randomizing seed control for fresh variety each queue.

Inputs and outputs that matter

• tags - your pool, multiline, split by the delimiter you choose.
• tags_delimiter - new line or , - pick whichever matches how you formatted the tags field.
• seed - controls the random draw.
• min_k / max_k - the range of how many tags get sampled, default 1/1 (exactly one). Bump max_k up for more variety per pull.

Output is a single STRING - the sampled tags, joined.

Installing it

ComfyUI Manager → search ComfyUI-PromptUtilities → install → restart. Or:

cd ComfyUI/custom_nodes && git clone https://github.com/nkchocoai/ComfyUI-PromptUtilities

Pure Python, no models or extra dependencies. It'll be sitting under the PromptUtilities category once you restart.

How this compares to the more established wildcard route

If you've spent any time in ComfyUI, you've probably already met wildcards through Impact Pack - its __wildcard__ file syntax and inline {a|b|c} random-choice syntax are what "a great many workflows depend on without crediting," as the community's own read on it puts it, and that's the far more established route for random tag pools. Sample Tags isn't trying to replace that. Its actual edge is the k-of-n range: pick anywhere from 1 to 4 tags per pull instead of committing to exactly one, and have the tag pool sitting as a visible textbox on the node rather than tucked away in a separate wildcard file you have to go find and edit.

Where people get burned

The main one: your delimiter has to actually match how you wrote the tags field. Pick , as the delimiter but write your list one-per-line, and you'll get one giant "tag" containing newlines instead of a clean split - or the reverse. Double-check the output on a small test list before trusting a big pull. Also worth testing: what happens if min_k exceeds how many tags you've actually got after splitting - the README doesn't document the failure mode, so don't assume it clamps gracefully, verify it on a short list first.