Upscale Model Loader (Searge)

This node loads an upscale model - an ESRGAN-family .pth file - so the rest of the Searge workflow can use it to enlarge images. You pick the upscaler by name, it loads it, and hands out a ready-to-use UPSCALE_MODEL. That's the whole job: it's the Searge equivalent of ComfyUI's stock "Load Upscale Model" node.

It's under Searge/_deprecated_/Files, part of the pack's older loader set. The Searge workflow's high-resolution and final-upscale passes need an actual upscaler model to run through, and this is where one gets pulled off disk.

How it works

You select a file from ComfyUI/models/upscale_models, and the node loads its weights into memory as an UPSCALE_MODEL object. Downstream, that model gets applied to an image to render it larger. ESRGAN-type upscalers are the workhorses here: they add pixels cleanly and near-instantly, and - importantly - they don't hallucinate new content. They make what's there bigger and sharper, not different. That's exactly what you want for the "more pixels" job as opposed to a generative restoration pass.

The inputs and outputs that matter

There's just one of each:

upscaler_name - the upscale model file to load.
UPSCALE_MODEL (output) - the loaded model, ready to feed an upscale step.

Which upscaler to grab

The Searge README recommends a specific set, and they're solid, community-favorite picks:

4x-UltraSharp - the general-purpose default; clean upscale without over-sharpening. This is the one most people reach for.
4x_NMKD-Siax_200k - good for anime and illustration.
4x_Nickelback_70000G - the README's suggested high-res upscaler.

Despite generative upscalers getting good, this plain-ESRGAN category hasn't gone anywhere - its share of upscaling discussion actually rose through 2026, because more people now know to fall back to a cheap, non-hallucinating upscaler when a generative one would invent the wrong details.

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: python -m pip install opencv-python in ComfyUI's Python env (required for the pack), then cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, restart. Windows portable has an installer script that can fetch the recommended upscalers for you.

You do need the upscaler files themselves. Download 4x-UltraSharp, 4x_NMKD-Siax_200k, and 4x_Nickelback (each ~67 MB) and drop them into ComfyUI/models/upscale_models. Without at least one file there, the dropdown is empty.

Where people get burned

Empty dropdown. No upscaler files installed means nothing to pick. Add .pth files to models/upscale_models and restart.
Expecting it to add detail. ESRGAN enlarges and cleans; it doesn't invent eyelashes and pores. For genuine restoration you'd want a generative upscaler (SUPIR, SeedVR2) - but those can also add unwanted texture to already-clean images. Match the tool to the job.
Wrong upscaler for the content. The anime-tuned models (NMKD-Siax) and photo-oriented ones behave differently. Using an anime upscaler on a photo, or vice versa, gives muddier results than picking the right one.