Image Resize To 🐦‍🔥

Most resizing nodes in ComfyUI make you pick a fixed width and height, which means you're deciding the aspect ratio too - often by accident, and always by squishing. This node takes the opposite approach: you tell it a target scale tier or two edge sizes, and it computes the other dimension automatically so the proportion is preserved. It's the "resize to 1080p" node, not the "resize to 1920×1080 even if that distorts" node.

How the presets work

The target_size dropdown offers 720P, 1080P, 2K, 4K, or custom. The presets aren't random - they're built around a 16:9 reference (720×1280, then scaled ×1.5, ×2, ×3), but the node then fits your image's own aspect ratio inside that box: whichever edge is longer drives the scale, and the short edge follows. A square image resized to "1080P" gets 1920×1920-ish behavior via the long-edge rule, not a forced 16:9 crop. That's the "proportion is kept by default" promise in the description.

Choose custom and the bbox_edge1 / bbox_edge2 inputs wake up - you specify two edge sizes (16–8192) and the orientation is picked automatically, with the same fit-inside-bounding-box logic. You don't have to say "which is the long edge"; it figures that out.

The other inputs that matter

• upscale_method: lanczos / nearest-exact / bilinear / area / bicubic. For upscaling, lanczos is the sharp default; area is notably good when downscaling. This is the same family of choices the core ComfyUI upscale nodes offer, because it calls ComfyUI's own common_upscale under the hood.
• onlyEnlarge (default true): the safety toggle. On (default), images larger than the target are left untouched - it will only ever upscale, never downscale. Turn it off if you genuinely want everything forced to the target size. This is the input that stops a batch job from shrinking your 4K masters by accident.

Outputs

Four of them, and the last two are the sleeper features: IMAGE (resized), prev_width / prev_height (the original dimensions, useful for logging or for a later restore), and log (a STRING that tells you whether it resized or skipped, and from what to what). In a batch pipeline, feeding that log into a print/save node gives you a free run log of what each image became.

Install

ComfyUI Manager → search ComfyUI-ArchiGraph, or:

cd ComfyUI/custom_nodes
git clone https://github.com/vincentfs/ComfyUI-ArchiGraph

Restart and run the pack's install script once. This node is pure torch/ComfyUI utilities - no OpenCV needed.

Gotchas

• Presets are 16:9-referenced. If you feed it an extreme panorama, "720P" won't give you a "720p video" frame - it gives a 720-long-edge image with your own ratio. Read it as a scale tier, not a literal resolution.
• onlyEnlarge defaults on. If your resizes "aren't working," that's usually why - the image was already bigger than the target.
• It upscales via common_upscale, which is a one-shot resample, not an AI upscaler. It'll make images bigger; it won't invent detail. For that you'd chain a real upscale model - this node is the geometry step.

The resize node you want when "keep my aspect ratio, make it 2K" is the whole requirement. Set the tier, leave onlyEnlarge on, done.