Edit Target Presets

Every image-edit prompt is two parts: what you're editing and what to do to it. This node is the first half - preset descriptions of the target. Pick HUMAN_FEMALE or MECHANICAL_OBJECT or INDOOR_SETTING, and out comes a edit_target_prompt string that describes that kind of subject or region in the instruction-style phrasing edit models actually respond to. It's the noun to UC_EditOpPresets' verb.

The list is organized by target kind and it's broader than it first looks. There are human targets (male, female, group, and more specific body-subject categories), anthropomorphic and robotic subjects, object classes like ADVERT_OBJECT (the default) and SEX_TOY_OBJECT, scene settings (indoor/outdoor), and property targets - lighting, shadows, skin, surface, texture. That last group is the quietly clever part: some edits aren't about a subject at all, they're about a property of the image, and having "texture" or "lighting" as a first-class target is the kind of thing you only add after writing your fiftieth "change the texture of the wall to..." prompt by hand.

The node is the standard preset shape: a preset dropdown, the escape_parentheses boolean ("prevents errors from parentheses left in the prompt"), and one edit_target_prompt string output. Wire it into a prompt concatenation with the operation preset and your own specific instruction. The intended grammar reads like a sentence: target + operation + your detail. "MECHANICAL_OBJECT + CHANGE_COMPOSITION + 'move the robot to the left side of the frame'" is a much more stable edit instruction than freehanding all three.

Why it works is the same reason the whole preset family works on modern encoders: they're instructions, and instructions with consistent vocabulary are more reliably followed. The KB's prompt-engineering work is blunt that this is the 2026 reality - your prompt is a message to a general-purpose LLM encoder, so structure and phrasing are the leverage. A preset target is fixed structure; it holds that part constant so your variable instruction stands out.

Caveats are shared with the sibling: presets describe categories, not this particular image. The target text tells the model what kind of thing to aim at, but your job is to tie it to the actual pixels - use a reference image socket or a detailed scene description, or the model may edit a "female subject" that isn't the one you meant. And property targets (texture, lighting) are lighter-weight than subject targets; if an edit model ignores one, it's usually because the property is genuinely ambiguous in the source image, not because the wording is wrong.

Install is pack-standard: Manager search "ComfyUI-UtilsCollection" or cd ComfyUI/custom_nodes && git clone https://github.com/silveroxides/ComfyUI-UtilsCollection then restart. No models, no downloads - it's a text node. Pair it with the edit-operation presets and you've got a small but real edit-prompt grammar kit.