Upscale Models Selector v2

This is the settings box where the Searge SDXL workflow picks its upscale models. Four dropdowns, four jobs, and like the rest of the Searge input nodes it doesn't upscale anything itself - it just records your choices into the data-stream bus so the Magic Box can use them during the high-res stages. Think of it as the control panel, not the machine.

The four slots map to the different upscaling roles the workflow uses across its passes: a primary upscaler, a secondary one, a high-res upscaler, and a detail processor. All four are pixel upscalers - the fast, non-generative ESRGAN kind - which is the right tool here. These add resolution to an image that already has the detail you want; they don't hallucinate new content and they run in milliseconds.

How it works

ESRGAN-family upscale models are trained to turn a small image into a larger one cleanly, without the softness of a plain resize. The Searge workflow runs more than one of them at different points, which is why there are four slots instead of one. The README recommends specific files for each role, and they're all community staples you'll recognise if you've done any SDXL upscaling.

The inputs and outputs that matter

Four required dropdowns, each reading from your models/upscale_models folder:

• primary_upscaler - the README suggests 4x-UltraSharp, the community's default general-purpose 4x model. Clean, no over-sharpening.
• secondary_upscaler - the README suggests 4x_NMKD-Siax_200k, a good all-rounder that's especially kind to illustration.
• high_res_upscaler - the README suggests 4x_Nickelback_70000G for the final high-res pass.
• detail_processor - optional; the README points at 1x-ITF-SkinDiffDetail-Lite-v1, a 1x model for skin detail rather than resolution. Leave it on none if you didn't download it.

The lone output is data (SRG_DATA_STREAM), chaining on to the next box or the Magic Box.

How to install it

Manager: search SeargeSDXL, install, restart. Manual install needs the opencv dependency first:

python -m pip install opencv-python
cd ComfyUI/custom_nodes
git clone https://github.com/SeargeDP/SeargeSDXL.git

Restart, then download the upscale models into ComfyUI/models/upscale_models. The README links all four from the uwg/upscaler HuggingFace repo (each of the 4x models is about 67 MB, the skin detailer 20 MB).

Common issues & troubleshooting

Dropdowns are empty. The node lists whatever's in models/upscale_models. If it's empty, you haven't downloaded any upscale models yet - grab the ones the README recommends and restart ComfyUI so they show up.

File-name mismatch. As with every Searge selector, the workflow ships with specific names pre-filled. If yours differ, pick them manually from each dropdown.

Don't overthink the model choice. These are all solid pixel upscalers and the differences between them are subtle on most images. The community consensus (per our upscaling notes) is that 4x-UltraSharp is a perfectly good default and this whole category is gaining share, not fading - a clean interpolating upscale is still the correct answer when your source already has the detail. If you want a model to invent detail on a soft or damaged image, that's a different job entirely (SeedVR2, SUPIR) and it isn't what these slots are for.