Nourivex Detailer Prompt XL

The name is a small lie. Nourivex Detailer Prompt XL doesn't detail anything - it doesn't detect a face, crop it, or re-render it. What it does is build the prompts you feed to whichever detailer you actually run. That's a genuinely useful job, because the prompt for a face-region re-render shouldn't be your main-pass prompt. Your base prompt says "full body, standing in a street, cinematic lighting." None of that helps a face pass that wants "sharp iris, clean facial structure." This node splits the difference for you.

How it works

It's a prompt assembler plus a CLIP encoder. You give it one base positive and one base negative - the prompts you'd use for the whole image - and it appends detailer-appropriate text to each, then encodes the results into conditioning. The assembly is dead simple: base_positive + face_style text + extra_face becomes the face positive, base_positive + body_style text + extra_body becomes the body positive, and the negatives get fixed anti-artifact strings tacked onto your base negative.

The style dropdowns are where the preset magic lives. face_style gives you four baked-in descriptions - Natural Face, Beauty Clean (the default), Anime Face, Sharp Photoreal - and body_style gives Natural Body, Athletic Definition, Soft Feminine, and a "Nuwa Mesh Consistency" option. Each one is just a pre-written chunk of descriptive text that gets pasted into the assembled prompt, so "Beauty Clean" resolves to something like "beauty shot face, symmetrical features, smooth but detailed skin, sharp iris." extra_face and extra_body are your free-text add-ons on top of that.

What wires where

Six outputs come out: four conditionings and two strings.

• face_positive / face_negative - feed these into your detailer's positive/negative inputs. In Impact Pack terms, that's the prompt slots on FaceDetailer or DetailerForEach (or any ADetailer-style pass, if you're wiring a manual inpaint crop).
• body_positive / body_negative - the optional second pass. The pack's own workflows treat body detailing as an extra that can be switched off, because a second full-body re-render is where pipelines get unstable and slow. Run it when the torso came out mushy; skip it when you just want the face fixed.
• face_prompt_text / body_prompt_text - the plain strings the node assembled before encoding. These are the underrated outputs: connect them to a display node and you can actually read what the face pass is being told, which is how you learn which style presets you actually like. They're also handy if you want to route the text into a different encoder.

The catch worth knowing

This is SDXL conditioning, built for the tag-and-weight prompting dialect. Everything here - the base defaults like "masterpiece, highres", the style text, the whole approach - assumes a CLIP-encoded SDXL model where commas and quality tags still do work. Point it at a Flux or LLM-encoded model and you're in the wrong dialect; the pack's own Flux workflow explicitly doesn't use this node for conditioning and encodes with CLIPTextEncodeFlux instead.

Also worth noting: the "Nuwa Mesh Consistency" body style refers to the author's own Nuwa LoRA series, and the README says the default workflows stopped shipping that LoRA because it caused bugs in some detailer scenarios. The preset will happily append its text whether or not you own the LoRA - it's just words, and it's only as good as the model you run it on.

Install

Same as the rest of the pack - ComfyUI Manager (search Nour_Comfy) or:

cd ComfyUI/custom_nodes
git clone https://github.com/Nourivex/Nour_Comfy

Restart ComfyUI, hard-refresh the browser if the nodes don't appear. No dependencies, no downloads.