ComfyUI Node
Qwen 3.5
A ComfyUI node in Qwen3.5 with 15 inputs and 2 outputs.
Qwen 3.5
- image
- video
- RESPONSE
- THINKING
◄modelQwen3.5-9B►
◄promptDescribe this image in detail.►
◄system_prompt►
◄max_tokens4096►
◄temperature1.00►
◄top_p0.95►
◄top_k20►
◄repetition_penalty1.00►
◄enable_thinkingtrue►
◄quantizationFP16►
◄keep_model_loadedtrue►
◄seed1►
◄frame_count16►
CategoryQwen3.5
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| model | COMBO | Qwen3.5-9B | Model size. 0.8B ~2GB, 2B ~5GB, 4B ~9GB, 9B ~20GB, 27B ~56GB (FP16) |
| prompt | STRING | Describe this image in detail. | Text prompt for the model |
| system_prompt | STRING | Optional system prompt to set model behavior | |
| max_tokens | INT | 409664–81920 | Maximum tokens to generate |
| temperature | FLOAT | 1.000–2 | Sampling temperature. Thinking mode recommends 1.0, instruct mode 0.7 |
| top_p | FLOAT | 0.950–1 | Nucleus sampling. Thinking mode recommends 0.95, instruct mode 0.8 |
| top_k | INT | 201–100 | Top-K sampling. Recommended: 20 |
| repetition_penalty | FLOAT | 1.000.5–2 | Penalty for repeated tokens. Recommended: 1.0 |
| enable_thinking | BOOLEAN | true | Enable thinking mode. Model outputs <think>...</think> reasoning before response. |
| quantization | COMBO | FP16 | Model quantization. 4-bit needs ~7GB VRAM, 8-bit ~12GB, FP16 ~20GB |
| keep_model_loaded | BOOLEAN | true | Keep model in VRAM between runs for faster inference |
| seed | INT | 11–4294967295 | Random seed for reproducibility |
| imageopt | IMAGE | Single image input | |
| videoopt | IMAGE | Video frames input (batch of images) | |
| frame_countopt | INT | 161–64 | Maximum number of frames to sample from video |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| RESPONSE | STRING | — |
| THINKING | STRING | — |