ComfyUI Node
QWEN General Prompt(Local)
General-purpose local Qwen3.8 multimodal generation: images plus a text prompt, with an optional Skill from the bundled or custom Skill directories.
QWEN General Prompt(Local)
- images
- videos
- text
- selected_skill
◄promptDescribe the connected images in detail.►
◄skillnone►
◄include_skill_referencestrue►
◄llm_modelQwen3.8-27B-Q4_K_M.gguf►
◄vision_modelmmproj-F16.gguf►
◄think_modefalse►
◄reasoning_effortmedium►
◄seed0►
◄max_tokens8192►
◄video_fps24.0►
◄video_sample_frames_per_sec2►
◄force_unload_modeltrue►
◄system_prompt►
Category😺dzNodes/Qwen_H3_Prompt
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Describe the connected images in detail. | The request sent to Qwen. Connected pictures follow it as visual input. |
| skill | COMBO | none | None answers from the prompt only. Auto routes across bundled and custom Skills; pick one id to load it directly. |
| include_skill_references | BOOLEAN | true | Off loads only SKILL.md and skips the Skill's references/ files, which saves context. |
| llm_model | COMBO | Qwen3.8-27B-Q4_K_M.gguf | GGUF language model from ComfyUI/models/LLM/Qwen3.8. |
| vision_model | COMBO | mmproj-F16.gguf | GGUF multimodal projector, or none for text-only runs without an mmproj file. |
| think_mode | BOOLEAN | false | Off uses Qwen's official instruct settings. On enables thinking and uses Qwen's official thinking settings. |
| reasoning_effort | COMBO | medium | Applied only in thinking mode. |
| seed | INT | 00–18446744073709550000 | ComfyUI seed. It is mapped deterministically to llama.cpp's 32-bit seed range. |
| max_tokens | INT | 8192256–16384 | Maximum generated tokens, including thinking when thinking mode is enabled. |
| video_fps | FLOAT | 24.00.5–120 | Frame rate of every connected video frame batch. It sets the timestamps shown to the model. |
| video_sample_frames_per_sec | INT | 21–8 | Frames sampled from each second of every video. Frames from the same second are sent as one chronological contact sheet. |
| force_unload_model | BOOLEAN | true | On stops the node's llama.cpp server and clears its VRAM and system memory after each run. Errors always force an unload. |
| system_promptopt | STRING | Extra standing instructions. Leave empty to rely on the prompt alone. | |
| imagesopt | COMFY_AUTOGROW_V3 | — | |
| videosopt | COMFY_AUTOGROW_V3 | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| text | STRING | — |
| selected_skill | STRING | — |