ComfyUI Node
Load Qwen3.5 CLIP (Anima)
A ComfyUI node in loaders/Anima with 5 inputs and 1 output.
Load Qwen3.5 CLIP (Anima)
- CLIP
◄clip_name▾►
◄use_calibrationfalse►
◄use_alignmentfalse►
◄alignment_strength0.00►
◄output_scale1.0►
Categoryloaders/Anima
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| clip_name | COMBO | 1 options: qwen35_4b.safetensors | |
| use_calibrationopt | BOOLEAN | false | Apply per-dimension affine calibration to align 4B output distribution with 0.6B. Requires calibration_params.safetensors (run calibrate.py to generate). |
| use_alignmentopt | BOOLEAN | false | Apply Procrustes rotation to align 4B spatial/pose concept directions with 0.6B. Requires rotation_matrix.safetensors (run compute_alignment.py to generate). Helps with 'from side', 'from behind', and other viewpoint tags. |
| alignment_strengthopt | FLOAT | 0.000–1 | Controls the bias shift strength. The rotation (fixing concept directions like 'from side') is always applied when alignment is on. This slider blends the distribution center: 0=keep 4B's own magnitude (L2~10), 1=shift to 0.6B's magnitude (L2~70). Try 0.0 first (rotation only) and increase if poses/viewpoints still need help. |
| output_scaleopt | FLOAT | 1.00–1000 | Additional uniform scale factor applied AFTER calibration (if enabled). Usually leave at 1.0 when calibration is on. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| CLIP | CLIP | — |