ImageMegapixelSelector

ImageMegapixelSelector resizes images to hit a target megapixel budget. It’s a quick way to keep batches inside memory limits or align images with downstream models that expect a specific pixel count.

Inputs

image – The photo to resize.
target_megapixels – Desired image size measured in megapixels (e.g., 1.5 equals 1.5 million pixels).
clamp_min_megapixels – Prevent downsizing below this value; useful when you only want to shrink oversize images.
round_dimensions – Choose how the node rounds width and height:
- Even keeps both dimensions even.
- Multiple of 64 suits latent models that require divisible sizes.
- None keeps the exact float result.
keep_aspect – When True, scale both sides evenly; when False, allow separate width/height scaling to hit the target precisely.

Outputs

image – Resized image tensor.
width, height – New dimensions in pixels.
scale_factor – Multiplier applied to the original size.

Where It Fits

Use the node before memory-heavy diffusion or upscaling steps when you need to guarantee each frame stays under a certain size. It’s also handy for standardising mixed-source reference sets so later processing behaves consistently.

Tuning Tips

Enable keep_aspect to avoid distortion; only disable it if you plan to crop afterward.
Choose Multiple of 64 rounding for Stable Diffusion checkpoints, or Even for general-purpose pipelines.
Set clamp_min_megapixels to the same value as target_megapixels when you only ever want to downscale oversized inputs.

Troubleshooting

Image still too big – Lower the target megapixels or check that keep_aspect isn’t preventing the exact resize you need.
Image looks stretched – Make sure keep_aspect is enabled unless non-uniform scaling is intentional.
Model rejects dimensions – Adjust the rounding mode to satisfy the model’s width/height requirements.

Screenshot: docs/screenshots/image_megapixel_selector.png

2.0 KiB Raw Blame History Unescape Escape