AI_ModelScope_text
Text chat via ModelScope, defaulting to MiniMax rather than Qwen
- AI回复
- status
The text-only chat node of the ModelScope trio, and the one detail that stands out is the model list itself: the default is MiniMax/MiniMax-M2.5, with ZhipuAI/GLM-5 as the other option - not a Qwen model at all, even though ModelScope is Alibaba's own hub. That's the useful thing to understand about ModelScope generally: it's a hosting platform, not just a showcase for Alibaba's in-house Tongyi models, so nodes built against it can reach other labs' models through the same API token and the same interface. If you specifically want Alibaba's own Qwen line for text chat, that's AI_Qwen_text in this same pack, hitting Alibaba's DashScope API directly instead of going through ModelScope.
Fields ship in Chinese, consistent with this pack's other ModelScope nodes: 预设模型 (preset model), 提示词预设 (prompt preset, defaults to "None" until you've saved one via AI_PresetSave), 自定义系统提示词 (custom system prompt), 用户消息 (user message - your actual prompt), 温度/temperature (0–2, default 0.7), Top-P (0–1, default 0.9), 最大长度/max length (1–32768, default 2048), 超时时间/timeout (1–600 seconds, default 180), seed, MODELSCOPE_API_TOKEN, and 自定义模型 for a free-text model override outside the two listed. There are no optional image inputs - this is text in, text out. Outputs are AI回复 (the reply) and status.
The 32768 ceiling on 最大长度 is notably higher than most of the other chat nodes in this pack (GLM and Qwen cap around 4096–8192), which makes this a reasonable pick if you're doing longer-form generation - summarizing a big block of text, drafting something substantial - rather than a short prompt-punch-up.
MiniMax as a default is also worth a second look if you've never used it: it's one of the labs that gets less attention in the local-generation community than Alibaba or Zhipu, but its models are a legitimate, competitive option for general text work rather than a filler default - don't assume "not one of the famous three" means "not worth trying."
Installing it: through ComfyUI Manager (search "ComfyUI-Apt_Preset") or manually - cd ComfyUI/custom_nodes && git clone https://github.com/cardenluo/ComfyUI-Apt_Preset - then install.bat and restart.
Getting a token: create an account at modelscope.cn and generate an API token from your profile settings. The free tier's quota is generous enough to actually test this node without paying first, which isn't true of every API node in this pack.
Troubleshooting: an empty or failing AI回复 is almost always MODELSCOPE_API_TOKEN - confirm it's valid before looking anywhere else. If 温度/temperature is set high and responses feel incoherent or off-topic, that's expected behavior for any LLM at high sampling temperature, not a fault in the node - pull it back toward the 0.7 default and it should stabilize. And if your saved preset from AI_PresetSave isn't showing up under 提示词预设, restart ComfyUI - this pack's presets consistently need a fresh load to register as selectable.
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| 预设模型 | COMBO | MiniMax/MiniMax-M2.5 | 3 options: None, MiniMax/MiniMax-M2.5, ZhipuAI/GLM-5 |
| 提示词预设 | COMBO | None | 1 options: None |
| 自定义系统提示词 | STRING | — | |
| 用户消息 | STRING | — | |
| 温度 | FLOAT | 0.700–2 | — |
| Top-P | FLOAT | 0.900–1 | — |
| 最大长度 | INT | 20481–32768 | — |
| 超时时间 | INT | 1801–600 | — |
| seed | INT | 00–2147483647 | — |
| MODELSCOPE_API_TOKEN | STRING | — | |
| 自定义模型 | STRING | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| AI回复 | STRING | — |
| status | STRING | — |