Resize Images by Shorter Edge (DEPRECATED)
Resize Images by Shorter Edge — the deprecated floor-setter
- images
- images
The mirror image of its more famous sibling. Where "Resize by Longer Edge" caps the maximum size, this node guarantees a minimum: it scales the image so the shorter edge hits your target, and the long edge comes out however it comes out. It's the "nothing smaller than this" resizer - and like its sibling, it's DEPRECATED, replaced by one node that does both.
What it does
Feed it images and a shorter_edge (default 512). It finds the short edge, scales the image so that edge equals your number, and preserves aspect ratio along the way. A 2048×1024 image with shorter_edge 512 becomes 1024×512 - the short edge (1024→512) drives it, and the long edge scales along. A square image just becomes 512×512.
Implementation notes, same as its sibling:
- LANCZOS resampling. This is pure interpolation - "more pixels," not "more detail." It won't invent anything.
- Batch-aware. Every image in the batch gets the same shorter-edge treatment, independently, so mixed-size stacks come out normalized to the same short edge.
Schema is minimal: images in, shorter_edge (1–8192, default 512) in, images out. One number, one job.
When it was the right call
"Set a floor on the short edge" is a real need. If a downstream model or step demands a minimum resolution on both axes - you want everything at least 512 on its smallest side - this is the direct expression of it. It was also a convenient way to normalize a batch of mostly-portrait or mostly-landscape images so they all clear a threshold before they hit a sampler.
But in practice it's the node people reach for when they think they want a target resolution and accidentally pick the short edge, then wonder why their images are wildly different sizes. The two deprecated resize nodes read like a pair of traps: one sets a ceiling, one sets a floor, and neither tells you the other dimension.
Why it's deprecated
ComfyUI's newer Resize Image/Mask node (ResizeImageMaskNode) has a "scale shorter dimension" mode that does exactly this, plus a dozen other modes - including "scale to total pixels" and "scale to multiple," which cover the actual use cases better. The source marks this node as superseded by that one, and it's flagged DEPRECATED in the node menu. It survives only in old saved workflows.
If you're resurrecting an old workflow and see it: that's what it's doing, and it's harmless to leave in place. If you're building new:
Replace "Resize by Shorter Edge" → Resize Image/Mask → resize_type: "scale shorter dimension"
You get the same behavior, plus masks resized in lockstep so image and mask never drift, plus the option to snap to multiples of 8 so the VAE doesn't crop your canvas later. That last one matters - a "shorter_edge 512" on an awkward aspect ratio produces odd long edges, and odd dimensions are exactly what the latent encoder will silently shave. With the new node, set it to scale-to-multiple and stop thinking about it.
One node replaced both of these, with better defaults and mask support. That's the entire story of this deprecation.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | Image to process. | |
| shorter_edge | INT | 5121–8192 | Target dimension for the shorter edge. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | Processed images |