ComfyUI Node
Eric Krea2 Control (Depth/Pose ControlNet-LoRA)
A ComfyUI node in Eric/Krea2 with 8 inputs and 2 outputs.
Eric Krea2 Control (Depth/Pose ControlNet-LoRA)
- control_image
- control
- control_preview
◄control_lora▾►
◄channel_modegrayscale►
◄normalizeper_image_minmax►
◄invertfalse►
◄s1_strength1.00►
◄s2_strength0.50►
◄s3_strength0.00►
CategoryEric/Krea2
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| control_image | IMAGE | Preprocessed control map (e.g. Depth Anything / DepthPro output for the depth LoRA). Any size - it is cover-cropped to each stage's resolution, so keep it at the generation's aspect ratio. | |
| control_lora | COMBO | Krea2 Control-LoRA checkpoint (expanded input layer + block LoRA). Files with 'control' in their path are listed first. | |
| channel_modeopt | COMBO | grayscale | grayscale: luminance repeated to RGB (depth maps). rgb: keep colors (pose / canny / normal / lineart LoRAs). |
| normalizeopt | COMBO | per_image_minmax | per_image_minmax stretches the map to 0..1 - matches the depth LoRA's training (inverse depth, min-max). Use none for pose/canny. |
| invertopt | BOOLEAN | false | Flip the map. The depth LoRA expects NEAR = WHITE; turn this on only if your depth preview shows near objects dark. |
| s1_strengthopt | FLOAT | 1.000–2 | Stage 1 control strength (= the reference's --lora-scale on the block LoRA; the expanded input layer is always full). 0 = stage runs the stock model. ~0.6 = weaker structure adherence, more freedom. |
| s2_strengthopt | FLOAT | 0.500–2 | Stage 2 strength. Structure is set in S1; a moderate value keeps the refine from drifting. 0 = stock model for this stage. |
| s3_strengthopt | FLOAT | 0.000–2 | Stage 3 strength (detail pass). Usually 0: the trained buckets are ~1 MP and S3 runs far above that. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| control | KREA2_CONTROL | — |
| control_preview | IMAGE | — |