Image scale by ratio

Sometimes you don't care about an exact target size - you just want "half this" or "twice this." That's what this node is for. Give it an image and a multiplier, and it scales both dimensions by that factor. 2.0 doubles it, 0.5 halves it, 1.5 gives you a 50% bump. The aspect ratio is preserved automatically because you're scaling both axes by the same number.

It's part of Derfuu_ComfyUI_ModdedNodes, an older utility pack whose whole reason to exist is automating this kind of size arithmetic inside the graph. If you want to pin an exact edge length instead of a ratio, its sibling Image scale to side is the one you want.

How it works

It's core ComfyUI-style resampling with a ratio input. The upscale_by number multiplies the current width and height; the node then resamples the pixels with the filter you pick. Nothing generative happens here - it's interpolation, the fast "more pixels" path. Scaling up won't invent detail that wasn't in the source.

The inputs that matter

• image (IMAGE) - the input.
• upscale_by (FLOAT, default 1.0) - the multiplier. Below 1 shrinks, above 1 grows. This is the one knob that does the work.
• upscale_method - the resampling filter: nearest-exact, bilinear, bicubic, bislerp, area, lanczos. For shrinking, area or lanczos. For growing, lanczos/bicubic. nearest-exact only for pixel art.
• crop - disabled or center, default disabled.

One output: IMAGE.

Where it earns its keep

The classic pairing is with a model upscaler. ImageUpscaleWithModel always applies its model's fixed factor - usually 4x - which is often more than you want. Run the 4x, then drop this after it at 0.5 and you've got a clean 2x net result with the detail benefit of the ESRGAN pass and none of the "why is my image 8000px" surprise. It's also handy for feeding a downscaled copy of an image into a second sampler pass.

Because upscale_by is a float input, you can also drive it from a math node - compute a ratio elsewhere in the graph and wire it in - which is exactly the kind of automation this pack was built for.

Common issues & troubleshooting

Node missing / red after an update. Derfuu has a habit of deleting old node classes when reorganizing the pack instead of leaving them deprecated, so updates can break saved workflows. Update via Manager and re-add the node if it disappears.

ModuleNotFoundError: No module named 'custom_nodes.Derfuu_ComfyUI_ModdedNodes'. An old graph pointing at a module path that moved during a reorg. Update the pack, reload, replace the stale node.

Blurry result when scaling up. That's interpolation doing its honest thing - there's no detail to recover. Use a real upscale model for sharpness and reserve this for sizing.

Odd final dimensions. A ratio on odd source sizes can produce non-even results, which some downstream nodes (VAE, samplers) dislike. If a later node complains, use Image scale to side to land on a clean, divisible number instead.

Install: ComfyUI Manager → search "Derfuu" → install → restart, or cd ComfyUI/custom_nodes && git clone https://github.com/Derfuu/Derfuu_ComfyUI_ModdedNodes then restart. No models, no extra pip dependencies.