肖像大师：风格 & 姿势

The other Portrait Master v3 modules answer "who is this person." This one answers "how is the photo taken." PortraitMasterStylePoseCN (肖像大师：风格 & 姿势) is the module that controls pose, wardrobe, lighting direction and type, two style slots, and a one-click photorealism booster. It's the difference between "a woman standing in a room" and "a woman leaning against a wall in evening light, shot on a real camera" - without you writing a single clause by hand.

It ships in comfyui-portrait-master-cn, the Chinese localization of Stefano Flore's Portrait Master. Pure prompt-builder; no models, no dependencies, no GPU.

How it works

Standard Portrait Master machinery: dropdown selectors resolve to English prompt phrases, weight sliders wrap them in (phrase:weight), - values get skipped, and 随机 🎲 picks a random entry. But this module has the two most interesting behaviors in the pack:

• 照片写实增强 (photorealism improvement) is a boolean here (in the legacy node it's a two-way dropdown). When on, it appends (professional photo, balanced photo, balanced exposure:1.2) to the prompt. That's the whole feature - and it's a good one, because that trio of tokens is exactly the lexical nudge that pushes SDXL-lineage checkpoints off the illustration plateau.
• 模特姿势 (model pose) pulls from a 90+ entry pose library (pose_list.txt). The README is explicit: disable ControlNet if you're using the pose library - the two fight each other. This is the rare case where a text-based pose list genuinely does the job, because each entry is a full descriptive phrase.

The inputs that matter

• 模特姿势 - pose from the library. Pick first, then frame with the shot in the BaseCharacter module.
• 服装 and 女士内衣 - 89 clothes options plus a 270-entry lingerie list (added after a Reddit thread on how Flux handles lingerie prompts). This is where a lot of the pack's vocabulary actually lives.
• 光照类型 + 光照方向 + 光照权重 - 70+ light types, 10 directions. This trio is criminally underrated; lighting direction is one of the highest-leverage things you can put in a portrait prompt.
• style_1 / style_2 + weights - 79 styles each (b&w photography, cinestill, golden hour…). Two weighted style slots is genuinely useful - stack "b&w photography" with "studio lighting" instead of hoping the model guesses.
• 妆容 here is a duplicate of the dedicated Makeup module's selector - harmless, just pick one place to set it.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/DropFan/comfyui-portrait-master-cn

Restart, or find comfyui-portrait-master-cn in ComfyUI Manager. Nothing to download beyond the clone.

Where people get burned

• Don't stack this module with ControlNet poses. The README calls it out directly: disable ControlNet when using 模特姿势, and adjust the framing with the shot selector instead. A ControlNet pose plus a text pose is how you get contradictory anatomy.
• The pose library reads like a caption, not a pose map. It's fantastic for the "casual, in-context" shots, but if you need a precise T-pose or a specific hand position, you still want ControlNet/OpenPose. This node is a style tool, not a skeleton rig.
• Weights assume a CLIP encoder. On LLM-encoder models (Flux, Z-Image) the (token:1.2) wrappers mostly get discarded; the vocabulary still helps, the weighting doesn't. Keep it for SDXL-lineage.

The honest take

If you can only install one module from this pack besides the BaseCharacter, make it this one. The light-direction + style-slot combo is the part of Portrait Master that survives the most model generations - it's teaching you the vocabulary of photography, and that knowledge transfers everywhere.