ComfyUI Node
Qwen3.5 Vision Encode (Anima)
A ComfyUI node in conditioning/Anima with 7 inputs and 1 output.
Qwen3.5 Vision Encode (Anima)
- clip
- vit
- image
- CONDITIONING
◄text—►
◄modeadd►
◄image_size448►
◄vision_weight0.30►
Categoryconditioning/Anima
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | — | |
| vit | QWEN35_VIT | — | |
| image | IMAGE | — | |
| text | STRING | Text prompt. Visual features from the image will be injected via the selected mode. | |
| modeopt | COMBO | add | How to inject visual features: - add: pool ViT patches into a style vector, add to every text token (best for style/character transfer) - replace_padding: overwrite padding tokens with visual features (middle ground) - concat: prepend all visual tokens before text (can drown out text prompt) |
| image_sizeopt | INT | 44864–768 | Resize image to this size (both H and W, must be multiple of 32). 448 = 196 visual tokens, 224 = 49 tokens. Larger = more detail but slower. |
| vision_weightopt | FLOAT | 0.300–10 | Scale factor for visual influence. For 'add' mode (pre-norm): 0.1=subtle, 0.3=moderate, 1.0=strong (visual same magnitude as text). For 'concat'/'replace_padding': scales projected token magnitudes. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| CONDITIONING | CONDITIONING | — |