ModelScope-LoRA 单LoRA加载

If you've used the ModelScope-Image or Image-Edit nodes, you've seen the lora1_id / lora1_w text fields sitting there empty. Those fields want a LoRA model ID as a string and a weight as a float, and typing model IDs by hand is how mistakes happen. This node exists so you don't have to: it turns the pack's saved LoRA presets into a dropdown and outputs the selected one's ID and weight, ready to wire in.

How it works

That's the whole job. No API call happens here - the node reads the lora_presets list from modelscope_config.json and hands back two values:

• lora_id (STRING) - the ModelScope LoRA model ID from the chosen preset, e.g. qiyuanai/TikTok_Xiaohongshu_career_line_beauty_v1.
• lora_weight (FLOAT) - the preset's stored weight, or your override.

The one input that matters is lora_preset, a dropdown built from the config's preset list (default "无LoRA", meaning no LoRA - an empty ID). The other two are the weight controls: lora_weight (0–2, default 0.8) is your value, and use_custom_weight (boolean, default off) decides whether your value gets used or whether the preset's stored weight wins. Leave it off and you get the preset's defaults; flip it on to tune one run without editing the preset.

Wiring it up

Wire it up like this: lora_id → the lora1_id input on a ModelScopeImageNode or ModelScopeImageEditNode, lora_weight → that node's lora1_w. Run the image node and the LoRA rides along in the API request's loras payload.

The preset list is the same one the pack ships with - the Chinese Xiaohongshu/Douyin-style LoRAs - so on a fresh install your dropdown options are those five plus 无LoRA. To add your own, use the ModelScopeLoraPresetNode, or edit modelscope_config.json directly. Either way, remember the dropdown is captured when the node is created, so new presets show up on fresh nodes (or after reloading the workflow), not retroactively.

Installing it

Installation is the standard pack routine: ComfyUI Manager, searching "ComfyUI ModelScope API Node", or

cd ComfyUI/custom_nodes
git clone https://github.com/hujuying/ComfyUI-ModelScope-API

then restart ComfyUI. No extra dependencies - it's pure data plumbing. If you only ever use one LoRA at a time, this is the node you want; the moment you need two or three stacked, the multi-loader version below is the same thing with three lanes.