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

Image Resize Universal - Universal Image Resizer

Node Purpose: Image Resize Universal performs unified resizing under multiple aspect ratio sources and fit modes, with complete and consistent mask output. It supports ratio presets, target-side control, resampling method selection, size multiple constraints, and multiple padding background strategies.

Inputs

Name	Port	Type	Default	Range	Description
image	optional	IMAGE	-	-	Input image batch; when absent, target size can be inferred via preset_ratio / get_image_size.
mask	optional	MASK	-	-	Input mask batch; when size matches the image, it is resized in sync; when absent, a default mask is generated.
get_image_size	optional	IMAGE	-	-	Reference image (first frame only) used to infer target size.
preset_ratio	-	COMBO	origin	origin / custom / 1:1 / 3:2 / 4:3 / 16:9 / 21:9 / 2:3 / 3:4 / 9:16 / 9:21	Aspect ratio source; origin uses input size, custom uses the proportional parameters below.
proportional_width	-	INT	1	1-8192	Ratio width in custom mode.
proportional_height	-	INT	1	1-8192	Ratio height in custom mode.
method	-	COMBO	lanczos	nearest / bilinear / lanczos / bicubic / hamming / box	Resampling method.
scale_to_side	-	COMBO	None	None / longest / shortest / width / height / length_to_sq_area	Target-side control; e.g., set the longest/shortest side or explicit width/height.
scale_to_length	-	INT	1024	1-8192	Target length for the selected scale_to_side mode.
fit	-	COMBO	crop	crop / pad / stretch	Fit mode: crop, pad, or stretch.
pad_color	-	STRING	1.0	grayscale/HEX/RGB/edge/average/extend/mirror	Padding background strategy; see “Padding Strategies”.
divisible_by	-	INT	8	1-1024	Ensure output width/height round up to a multiple of this value (often required by models).

Outputs

Name	Type	Description
image	IMAGE	Resized image batch; can generate solid background images in no-image scenarios.
mask	MASK	Mask batch matched to output size; generated according to rules below.

Features

• Ratio inference: supports original size, standard ratios, and custom ratios; combined with target-side control to compute target width/height.
• Size constraints: divisible_by ensures the output dimensions are integer multiples of the given value.
• Resampling methods: nearest, bilinear, lanczos, bicubic, hamming, box to balance quality and performance.
• Fit modes:
  • crop: keep aspect, then center-crop to target size.
  • pad: keep aspect, then center-pad to target size; strictly distinguishes original and padded areas.
  • stretch: directly stretch to target size.
• No-image input: infer target size via preset_ratio / scale_to_side or get_image_size, and generate a background image.
• Mask generation: outputs meaningful masks for all valid input combinations; synchronized resizing when a mask is provided.

Mask Rules

• pad mode: output mask matches target size; original image area is white (255), padded area is black (0).
• crop mode: mask expresses the cropped region (white) based on the original, then converted to output.
• stretch mode: mask matches the output size and is fully white (255).
• Default masks: when no input mask is provided, a corresponding mask is generated per output image.

Padding Strategies (pad_color)

• Grayscale value: e.g., 0.5 indicates 50% gray and is converted to RGB.
• HEX: e.g., #FF0000 or FF0000.
• RGB: e.g., 255,0,0 or 0.5,0.2,0.8 (0–1 values are auto-converted).
• edge: fill with average edge color (top/bottom or left/right depending on orientation).
• average: fill with the global average color of the image.
• extend: replicate edge pixels (replicate).
• mirror: reflect edge pixels (reflect, with segmented extension support).

Typical Usage

• Preserve aspect while controlling the longest side: set preset_ratio=origin, scale_to_side=longest, and scale_to_length as needed.
• Align inputs to model constraints: set fit=pad, and divisible_by=8/16 to meet model size alignment.
• Batch resize with synchronized masks: connect both image and mask so the node resizes them consistently and outputs corresponding batches.

Notes & Tips

• Matching image and mask sizes yields the most stable synchronized resizing; providing only masks can also serve as a size reference.
• In no-image scenarios, provide get_image_size or explicit ratio/target-side settings to ensure deterministic output size.