SEGS Pick

Impact Pack's SEGS is a batch of detected regions - faces, objects, whatever your detector found - all bundled into one wire. Most of the time you hand that batch to a Detailer (SEGS) node and it processes everything. But sometimes you want just one of those segments, on its own, to feed into standard ComfyUI nodes that don't know SEGS from a hole in the ground. XJSegsPick takes a SEGS stream and an index and returns a single SEG. It's the "index into the array" node, and it unlocks the whole family of Extract and Stitch nodes this pack builds around.

Under the hood it's refreshingly honest: it pulls the valid segment list (same filtering as SEGS Count - empty masks and degenerate bboxes are skipped), checks your index is in range, and hands back seg_list[index]. The index is zero-based and clamps nothing: out of range means a hard error with a message telling you the valid range, which is actually the right behavior for a workflow bug you'd rather catch loudly than let silently corrupt.

The interface:

• segs - SEGS in.
• index - INT, zero-based, from 0 up to 10000.
• Output: seg - a single SEG.

Where it shines: the manual detailer pipeline. Pick a segment → run it through SEGS Extractor to get the image/mask/coordinates as normal types → process them in whatever nodes you want (a different sampler, an upscaler, an IPAdapter) → put it back with SEGS Stitcher. Impact Pack's Detailer is a great one-node black box, but the moment you want a different pipeline per region, Pick is how you stop working on the whole batch. Pair it with SEGS Count or a loop index and you can iterate a batch one segment at a time with full control over each pass.

Install is the shared pack routine:

cd ComfyUI/custom_nodes
git clone https://github.com/alexjx/ComfyUI-XJNodes

Restart ComfyUI and find it under XJNodes/segs, or use ComfyUI Manager and search "ComfyUI-XJNodes". No models to fetch, requirements.txt is empty - the only real requirement is that SEGS comes from Impact Pack, so it assumes you're running that ecosystem already.

Caveat with the personal-use territory: this pack has essentially no community footprint, so you're on your own for examples. The pattern itself - pick, extract, process, stitch - is standard SEGS work, and this node is the cleanest way to start that loop.