Edit Target Presets (Legacy)

Editing models have a failure mode that's maddening: they know how to change a thing but drift on what you pointed at. The classic fix, and the one the good workflows use, is to explicitly frame the target - "the edit focuses on X, keep everything else alone." Edit Target Presets (Legacy) is that framing, pre-written. Pick a target from the dropdown and you get a paragraph that tells the model exactly what region of the image the edit concerns.

Mechanically it's the same template-lookup-and-escape design as the other preset nodes in the pack: a dropdown into a collection of EDIT_TARGET_* strings, with the escape_parentheses guard. The targets are concrete and practical - GENERAL_SUBJECT, HUMAN_FEMALE, HUMAN_MALE, HUMAN_GROUP, ANTHRO_FEMALE, ANTHRO_MALE, and so on - and each template is written in the "the edit focuses on… maintain visual coherence… keep the unedited parts in focus" register that instruction-edit models respond to. It's the subject half of an edit instruction; pair it with an operation.

The inputs that matter

• preset - which subject the edit should target. Pick the one that matches what's in your image.
• escape_parentheses - off by default.

The output

edit_target_prompt - a STRING that names the edit subject. It slots into an edit workflow as the "what" half: combine it with an operation prompt (see Edit Operation Presets, the natural partner) and feed the combined instruction to your edit encoder.

Installing it

Part of silveroxides/ComfyUI-UtilsCollection. Install via ComfyUI Manager (search ComfyUI-UtilsCollection) or:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-UtilsCollection

Restart after. Dependencies: opencv-python, typing-extensions. Templates ship in the pack; nothing to download. "(Legacy)" marks the compat alias for the pack's canonical UC_EditTargetPresets.

Where people get burned

A target alone is not an instruction. Feeding just "the edit focuses on the female subject…" gives the model a subject and no action - it doesn't know what to change. The intended use is target + operation. And because these templates name a subject explicitly, mismatch is the failure: tell the model the target is a human female and it will spend effort preserving a subject that isn't there. Match the target to what's actually in the frame, or the "keep the unedited parts in focus" language gets wasted on the wrong region.