模特姿势更改

Once you've got one good shot of a model in your garment (say from ModelGenNode), you almost immediately want the same look in different poses - arms crossed, turned to the side, hands in pockets. Photographing all of that costs real money. This node, 模特姿势更改 ("model pose change"), is the lazy way out: one source image in, new poses out, same person, same clothes.

The mechanism is the pack standard: your image_input is base64-encoded and POSTed to Mojie's backend under the moter-pose-change model, and the returned image (or images) are downloaded back into the graph. Nothing runs locally - you're renting a hosted model's understanding of how a body can shift without ripping the fabric, which is a genuinely hard thing to get right and exactly why nobody's happy running this class of edit on a low-end GPU at home.

The controls that matter

• style - three modes. basic is the safe default, deep leans into a more dramatic restyle, and prompt hands the whole thing to your text. Pick prompt and the node requires a non-empty prompt - it'll throw an error otherwise, and rightly so, since you're asking the model to invent the pose from words alone.
• out_batch - 1 to 4. This is the multiplier: set it to 4 and you get four pose variants back in one call, stacked into a batch tensor. Cheaper than four separate runs on your balance, and it's the whole reason the node sits well in a sheet-generation workflow.
• resolution - 1K or 2K. Output only, but it changes the per-call cost.
• seed - normal diffusion seed. Note the source offsets it by a fixed amount before sending, so don't expect the exact seed you type - just expect that changing it changes results.

Output is output, an IMAGE batch (batch size = whatever out_batch produced). Split it with a BatchIndex/SplitBatch-style node if you need individual frames, or feed the whole batch into the pack's image-editor nodes.

Installing it

Same pack, same ritual: ComfyUI Manager search "mojieapi_party", or

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

then put your key from mojieaigc.com into config.ini and restart. requirements.txt is small (requests, aiohttp, toml, opencv-python) and there are zero model files to track down. The pack ships a prebuilt workflow, 服装模特生成(合集).json, that chains this node with model generation and clothes replacement - drag it in to see the intended wiring.

Common issues

The usual red-error-image channel applies: 401 = key problem, 403 = balance. If you select style = prompt with an empty prompt, the node raises a ValueError before anything is sent - that's your signal, not a crash. And because pose changes are one of the more expensive hosted operations in the pack, watch your balance; a multi-batch run multiplies the credit burn. If results look "samey" across runs, that's the fixed seed offset working against you - nudge the seed and it'll move.