easy portraitMaster

Instead of remembering the magic words for pores and catchlights and skin imperfections, easy portraitMaster gives you a wall of dropdowns and sliders that assemble a portrait prompt for you. It's Easy-Use's port of florestefano1975's well-known Portrait Master node, and it's aimed squarely at people who want photoreal-ish faces without having learned the whole vocabulary of portrait prompting. Set the shot, gender, age, nationality, expression, hair; nudge a few realism sliders; out comes a positive and negative prompt.

It's a nice on-ramp, and it's honest about what it is: a very elaborate prompt template. Worth calibrating your expectations up front, though - the whole approach (weighted tokens like skin details:1.4) is native to the SD 1.5 and SDXL era it was built in. On modern language-model-encoder models (Flux, Z-Image, Qwen), that kind of quality-token weighting does a lot less, and sometimes nothing. If you're on SDXL or a 1.5 realism checkpoint, this node is right at home. On the newest bases, treat its output as a starting draft you then rewrite in plain photographic language.

How it works

Every widget maps to a chunk of prompt text with an associated weight. The dropdowns (shot, gender, nationality, hair style, light type…) drop in descriptive phrases; the 0–1.95 sliders (skin_details, skin_pores, freckles, eyes_details, iris_details, and so on) emit weighted tokens at the strength you dial. It stitches all of that together - with your prompt_start, prompt_additional, and prompt_end text wrapped around it - into a finished positive prompt, plus a matching negative. Nothing is generated here; you get two strings to encode.

The inputs and outputs that matter

There are dozens of controls; a beginner really only needs a handful:

• shot and shot_weight - framing (headshot, close-up, etc.) and how hard to push it.
• gender, age, nationality_1 (and nationality_2 + nationality_mix to blend two) - the core identity of the subject.
• facial_expression, hair_style, hair_color, eyes_color - the obvious descriptors.
• The realism sliders - skin_details, skin_pores, and friends. This is where the "does it look like a photo or a plastic doll" lever lives. Per the photorealism playbook, too-smooth skin is the number-one tell that gives away an AI face, so a touch of skin_details / skin_pores is what pushes it toward believable. Don't crank everything to max - that overcooks fast.
• prompt_start (defaults to raw photo, (realistic:1.5)), prompt_additional, prompt_end, negative_prompt - your own text woven around the generated parts.

Outputs: positive and negative, both STRING. Encode them like any other prompt.

How to install it

Ships in ComfyUI-Easy-Use. ComfyUI Manager: search ComfyUI-Easy-Use, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

then install.bat (Windows) or pip install -r ComfyUI-Easy-Use/requirements.txt, and restart. No model downloads - it's a text builder.

Common issues & troubleshooting

Everything looks over-baked. Leaving a pile of sliders at high values stacks a lot of weighted tokens, and the result fights itself - plastic skin, exaggerated features. Start with most sliders near zero and raise only the two or three you care about (usually the skin ones). A focused prompt beats a maximal one.

Weak or no effect on Flux / Z-Image / Qwen. As noted up top, this is SD-era prompt engineering. The newer language-model encoders largely ignore (token:1.4) weighting, and quality-booster spam is close to inert on them. On those models, use portraitMaster's phrasing as inspiration but write like a photographer - name a camera, a lens, a film stock - which is what actually moves those models.

Output isn't conditioning. Same caveat as the other prompt nodes: positive and negative are strings. Send them into an encode / loader prompt input, not straight into a sampler.

Nationality/ethnicity terms feel blunt. They're literal prompt phrases and behave like any other token. If a result skews wrong, adjust the nationality_mix blend or override the relevant part via prompt_additional rather than expecting the dropdown alone to nail it.