服装模特生成
Turn a flat garment photo into a model wearing it
- cloths_image
- face_image
- output
You have a photo of a garment on a hanger (or laid flat on a table) and no model to put it on. A photoshoot is expensive, a stock model is a coin flip on whether the fit looks right, and your listing needs a body wearing the thing. That's the exact gap this node is built for: drop in the garment, get out a model wearing it. It's the front door of the pack's clothing series - 服装模特生成, "fashion model generation."
The hook is that everything about the model is your choice before you click. You pick ethnicity (race_class: 亚裔 / 黑人 / 白人), gender and age (gender_class: man / woman / little boy / little girl), a style preset (style_prompt - eight Chinese presets like 通用-INS自拍, an "Instagram selfie" vibe, or 女装-清新室内, "fresh indoor"), output Size (1:1, 3:4, 9:16) and resolution (1K or 2K). If you want a specific face on that body - say, you have a returning customer and want continuity - hook face_image in and the model will use it. That optional prompt field is your escape hatch when a preset doesn't fit; leave it empty and the presets carry the load.
How it works
Mechanically it's the same story as every node in this pack: your images are base64-encoded and POSTed to Mojie's endpoint under the mojie-output-moter model, and the result is downloaded and returned as an IMAGE tensor. The Chinese race_class values get mapped to the backend's Asia / black / Ukraine codes (yes, "Ukraine" is what the backend calls its white-person class - don't ask), and is_face flips to true the moment a face image is connected. Under the hood it's a hosted model doing the whole try-on; you are not running a diffusion model locally, which is the point if your machine is modest. This is the exact "the model isn't on your machine" pattern the KB's API-nodes essay covers - you trade local control for a capability a low-end PC can't reach.
The inputs that matter
cloths_image- the garment photo. Required, and it's the one thing you can't skip.face_image- optional, but if you want the model's face to match someone specific, this is where it goes.style_prompt+gender_class+race_class- the dials that stop you from generating a lineup of identical-looking models.seed- normal diffusion seed, change it for variety.
Output is output, a single IMAGE you can preview, save, or feed into the pack's other clothing nodes.
Installing and using it
Same as the rest of the pack - search "mojieapi_party" in ComfyUI Manager, or git clone the repo into custom_nodes, drop your key into config.ini from mojieaigc.com, restart. No model files to download. The author ships ready-made workflows in the pack's workflow/ folder - 服装模特生成(合集).json is the four-feature combo (model gen, white-background extraction, pose change, clothes swap) and the (生成组图).json one chains them into a sheet. Drag one in and you'll see how this node is meant to sit upstream of MoterPoseNode and ReplaceClothesNode.
Where people get burned
The error channel is a red image with the message painted on it. A 401 or 403 almost always means the key is wrong or the balance ran out - these calls are metered, and model generation eats credits a few cents at a time. And one culture note: this is a Chinese-market pack, so the prompts and presets are Chinese-first. You can pass English in the optional prompt, but if you want the presets' full effect, reading a little Chinese goes a long way.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| cloths_image | IMAGE | — | |
| race_class | COMBO | 亚裔 | 3 options: 亚裔, 黑人, 白人 |
| resolution | COMBO | 2K | 2 options: 1K, 2K |
| gender_class | COMBO | woman | 4 options: man, woman, little boy, little girl |
| style_prompt | COMBO | 通用-INS自拍 | 8 options: 通用-INS自拍, 女装-涉谷街拍, 通用-简约风, 女装-清新室内, 通用-靠墙特写, 通用-露营风, +2 |
| seed | INT | 0 | — |
| Size | COMBO | 3:4 | 3 options: 1:1, 3:4, 9:16 |
| face_imageopt | IMAGE | — | |
| promptopt | STRING | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| output | IMAGE | — |