ComfyUI Node
Eric Qwen Direction Apply
A ComfyUI node in Eric Qwen-Edit/Conditioning/Directions with 9 inputs and 1 output.
Eric Qwen Direction Apply
- conditioning
- direction
- conditioning
◄scale1.00►
◄methodlerp►
◄scopeall_tokens►
◄prefix_N0►
◄scope_N7►
◄normalize_directionfalse►
◄target_rms2.0►
CategoryEric Qwen-Edit/Conditioning/Directions
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| conditioning | QWEN_CONDITIONING | Base conditioning to modify | |
| direction | QWEN_DIRECTION | Direction to apply | |
| scale | FLOAT | 1.00-5–5 | 0=passthrough, 1=standard, >1=amplified, <0=opposite. With a trained LoRA, start at 0.5 - LoRA increases responsiveness. |
| methodopt | COMBO | lerp | lerp: linear interpolation (fast, good for scale 0.3-1.2). slerp: spherical interpolation (preserves embedding magnitude; better for large scale >1.5 or negative scale). |
| scopeopt | COMBO | all_tokens | all_tokens: apply to entire sequence. skip_prefix_N: text tokens only → expression / attribute. keep_prefix_N: image tokens only → style / film / visual. last_N_tokens: not recommended for Qwen. |
| prefix_Nopt | INT | 00–512 | 0 = auto from Encode metadata. |
| scope_Nopt | INT | 71–128 | Only used for last_N_tokens scope. |
| normalize_directionopt | BOOLEAN | false | Normalize direction to target_rms before applying scale. Makes scale comparable across direction files. Recommended when chaining multiple directions. |
| target_rmsopt | FLOAT | 2.00.1–10 | RMS target when normalize_direction=True. Use same value across all Apply nodes. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| conditioning | QWEN_CONDITIONING | — |