ComfyUI Node
Load LLM as CLIP
A ComfyUI node in LLM/text encoder with 7 inputs and 1 output.
Load LLM as CLIP
- clip
◄model_folder▾►
◄system_promptYou are an assistant designed to generate high-quality images with the highest degree of image-text alignment based on textual prompts.►
◄devicecuda►
◄dtypeFP16►
◄target_hidden_size2304►
◄enable_weightstrue►
◄force_offloadtrue►
CategoryLLM/text encoder
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| model_folder | COMBO | 1 options: (请将模型放到 models/LLM 目录) | |
| system_prompt | STRING | You are an assistant designed to generate high-quality images with the highest degree of image-text alignment based on textual prompts. | — |
| device | COMBO | cuda | 3 options: auto, cpu, cuda |
| dtype | COMBO | FP16 | 6 options: default, BF16, FP32, FP16, FP8_E4M3, FP8_E5M2 |
| target_hidden_size | INT | 2304512–8192 | 目标隐藏层大小。Gemma-2 2B=2304, Qwen-3=1024 |
| enable_weights | BOOLEAN | true | 启用权重功能：支持 (word:1.5) 语法和超长提示词自动分批 |
| force_offload | BOOLEAN | true | 强制模型在不使用时卸载到CPU，减少显存占用 |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| clip | CLIP | — |