BMAB Upscale With Model

What it is

This is BMAB's take on the classic ESRGAN-class upscaler - the cheap, reliable rung of upscaling where you're adding pixels, not inventing detail. No hallucinated skin texture, no risk of the model quietly redrawing a face, just a trained super-resolution model scaling the image up. If you already know what you're doing with upscale models, this behaves exactly the way you'd expect; the interesting part is that it plugs directly into BMAB's bind system, so it can slot into a larger BMAB pipeline without needing every input rewired by hand.

Reach for this over a generative upscaler (SeedVR2, SUPIR, that kind of thing) when the source is already reasonably clean and you just need more pixels - it's near-instant and costs almost nothing in VRAM by comparison. If the image actually needs detail rebuilt rather than just resized, this isn't the tool; that's a different, heavier job.

Inputs and outputs

model_name is a dropdown of whatever ESRGAN-style upscale models you've got installed - populated from ComfyUI's models/upscale_models folder, same convention as ComfyUI's native upscale node. scale (default 2, range 0–4, with a fine 0.001 step) is your multiplier. width and height (both default 512, up to 16384, step 8) let you cap or resize the final output to a specific target size after the model upscale runs, rather than being stuck with whatever the raw scale factor produces.

Two optional inputs: bind, a BMAB bind from ToBind or elsewhere in the pack, and image, the picture to upscale. Two outputs mirror that: BMAB bind passed through so you can keep chaining more BMAB nodes after this one without rewiring, and image, the upscaled result.

How to install it

Search comfyui_bmab in ComfyUI Manager - it'll install the pack along with the two dependencies the wider BMAB ecosystem needs, comfyui_controlnet_aux and ComfyUI_IPAdapter_plus. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/portu-sim/comfyui_bmab.git
cd comfyui_bmab
pip install -r requirements.txt

then clone and pip-install the two companion repos the same way, and restart. This node itself needs an upscale model separately - BMAB doesn't ship one - so drop an ESRGAN-family .pth (4x-UltraSharp is a solid general-purpose default) into ComfyUI/models/upscale_models before expecting the dropdown to show anything useful.

Common issues

Empty model_name dropdown means exactly what it sounds like: no upscale model installed yet. Grab one and restart ComfyUI, or at least refresh the node.

Watch the combination of a high scale value with a large source image - this node doesn't appear to expose any tiling option, so it's doing the upscale on the full image in one pass. On a big image at 4x that can be a genuinely heavy VRAM hit with no built-in way to chunk it down; if you hit an out-of-memory error, dropping the scale or pre-shrinking the source is your fastest fix, since there's no tile-size knob here to reach for instead.

And don't expect this node to fix anything that's already wrong with the image - faces, hands, artifacts, all of that survives the upscale untouched, just bigger. If the source has defects, handle those before this node runs, not after; upscaling a flawed image just bakes the flaw in at higher resolution.