1hew-ComfyUI-1hewNodes/web/docs/1hew_ImageHLFreqTransform/en.md

Image HL Freq Transform - Masked Detail Transfer

Node Purpose: Image HL Freq Transform transfers detail from a detail_image onto a generate_image using rgb, hsv, or igbi separation methods, with optional detail_mask to control where the detail applies. Outputs the result and the intermediate high/low layers.

Inputs

Name	Port	Type	Default	Range	Description
generate_image	-	IMAGE	-	-	Base image to receive detail.
detail_image	-	IMAGE	-	-	Source image providing detail.
detail_mask	optional	MASK	-	-	Mask controlling per-pixel detail transfer; full-white when absent.
method	-	COMBO	igbi	rgb / hsv / igbi	Separation method for transfer.
blur_radius	-	FLOAT	10.0	0.0–100.0	Gaussian blur radius; ensured odd and ≥3.

Outputs

Name	Type	Description
image	IMAGE	Result image with transferred detail.
high_freq	IMAGE	Intermediate high-frequency layer (after mixing).
low_freq	IMAGE	Intermediate low-frequency layer.

Features

• Batch expansion: repeats smaller batches across generate_image, detail_image, and detail_mask to match the largest.
• Mask semantics: when absent, a full-white 3-channel mask is used; otherwise it is broadcast to 3 channels.
• Methods:
• igbi transfer: builds high/low via invert+blur mixes from both images, blends highs by the mask, then mixes 0.65*high + 0.35*low and applies levels.
• rgb/hsv transfer: separates high/low from both images, blends highs by the mask, and recombines via linear light or HSV.
• Odd radius: blur radius normalized to an odd integer.

Typical Usage

• Texture from reference: bring fine texture from a reference into a generated base while preserving base tone in low-frequency.
• Masked retouch: apply detail only in selected regions with detail_mask (e.g., skin pores, fabric).
• Method selection: choose hsv for brightness-only transfer, rgb for direct contrast transfer, igbi for level-managed stylization.

Notes & Tips

• Inputs are converted to NumPy for OpenCV operations; outputs are in [0,1] float32.
• Large blur_radius widens the low-frequency envelope and softens high-frequency extraction.