人设设计

人设设计, "character design," is the odd one out in this pack's Product & Tool menu - most of those nodes are e-commerce plumbing, but this one is for the artists. You feed it a character reference (or a stack of them), pick a sheet type, and get back a proper character design sheet: the single reference portrait, a multi-angle turnaround, or a multi-expression board. If you've ever hand-assembled a character sheet in an image editor because your local model kept drifting on identity, this is the node that does the assembly for you - on Mojie's hosted human_desgin model.

Character consistency is the eternal problem here, and the KB's character-consistency essay is worth a skim before you blame the node: keeping the same face across poses and expressions is the hardest thing in this whole field, and no hosted model is immune to it. What this node gives you is a tool designed around the constraint - the multi-angle and multi-expression modes are explicitly built to hold one character while varying the view.

Modes and their rules

The type enum decides everything:

• 单张设定图 (single sheet) - your main reference portrait. Takes up to 10 input images, which is handy if you're building from multiple views of an existing character.
• 多角度视图 (multi-angle) - a turnaround. Strictly one input image; the model expands it into multiple angles.
• 多表情视图 (multi-expression) - a full expression board. Also one input image, and it requires a prompt describing the expressions you want. Leave it empty and the node raises a ValueError before it ever spends a credit - a rare example of this pack failing loud and early, which I appreciate.

The prompt field is optional on the other two modes. seed is your usual variety dial.

Inputs and output

image_input (required), type, seed, and optional prompt. Output is output - and note it can be multiple images: the backend returns several URLs joined by |, and the node splits them into a batch tensor. So a multi-expression run comes back as a batch you can preview together or split apart.

Installing it

Same pack, same two-line ritual:

cd ComfyUI/custom_nodes
git clone https://github.com/MoJIeAIGC/comfyui-MJAPI-party.git

or search "mojieapi_party" in ComfyUI Manager, then drop your key from mojieaigc.com into config.ini and restart. No model files, minimal dependencies, and it needs the key even though it looks like a local art tool - every image leaves the machine and the calls are metered.

Where people get stuck

• "多表情视图需传入提示词" - this exact error means you picked multi-expression without a prompt. Describe the expressions: "happy, angry, surprised, crying" style works fine.
• "最多只能输入1张图片" - the multi-angle and multi-expression modes hard-cap at one input image. Feed them a batch and they'll refuse.
• Face drift across expressions. When expressions warp the identity, that's the underlying model's limit, not a wiring mistake. A clearer single reference image is your best lever; garbage in, character-sheet garbage out.
• Cost. Character design calls are hosted and metered, and expression boards are multi-image jobs, so they burn more than a single-sheet run. The red error image is the pack's way of announcing an empty balance.