SP_HiresGen_HiresCfg
Which upscale model and how many passes for SP_HiresGen
- hires_cfg
The smallest of the three config nodes that feed SP_HiresGen, and the most literal: it just answers two questions - which upscale model, and how many times to run the hi-res loop. Everything about how each of those iterations behaves (resolution, steps, CFG, denoise) lives in the sibling node SP_HiresGen_Dynamic instead.
How it works
model is a dropdown populated from whatever's sitting in your models/upscale_models folder, defaulting to 4x-ultrasharp.pth. UltraSharp is one of the better-known standalone ESRGAN-family upscalers in the community - the kind of model people reach for as the "apply after generation" step in the classic ESRGAN upscale workflow, separate from and complementary to the img2img-based hi-res fix pass. SP_HiresGen uses it as the resize step between each of its generative iterations, so the choice of model here shapes the texture of every pass, not just a final polish.
iterations sets how many times SP_HiresGen loops through generate-and-upscale before it's done - default 3. This number is what SP_HiresGen_Dynamic's start/end ramps are stretched across: three iterations means three points sampled along that ramp, more iterations means finer, smaller steps between the start and end values you configured there.
Inputs and outputs that matter
model- the upscale model, default4x-ultrasharp.pth. Swap it for whatever you have installed; the choices list is populated from your actualupscale_modelsfolder, so if it's empty here, that's a sign the folder is empty too.iterations(default 3, 1-10000 range) - how many hi-res passes to run. Realistically you'll stay in the 2-5 range; higher values mostly mean longer render times for diminishing returns once you're past your target resolution.- Output:
hires_cfg, feeds directly into SP_HiresGen's requiredhires_cfginput.
How to install it
Bundled with the rest of SP-Nodes. Through ComfyUI Manager: search "SP-Nodes," install, restart. Manually:
cd ComfyUI/custom_nodes
git clone https://github.com/bananasss00/ComfyUI-SP-Nodes
Then restart ComfyUI. The node itself needs nothing extra, but its default model choice does: 4x-ultrasharp.pth has to actually be downloaded and placed in models/upscale_models - it isn't bundled with the pack, and the pack's README doesn't link it, so grab it from OpenModelDB or wherever you source your other ESRGAN models.
Common issues & troubleshooting
The model dropdown is empty or missing your expected default. That means ComfyUI can't see any upscale models in models/upscale_models - download at least one (UltraSharp is the sane starting point) and restart ComfyUI so it re-scans the folder.
Runs take much longer than expected. Check iterations - it's easy to bump this up while experimenting and forget it multiplies your total render time roughly linearly, since each iteration is a full generate-and-upscale pass, not a cheap incremental step.
Detail plateaus after a couple of passes. That's expected once you've reached a resolution where the upscale model isn't adding meaningfully more than the previous pass - more iterations past that point mostly cost time, not quality; SP_HiresGen_Dynamic's end values matter more than raising iterations further.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| model | COMBO | 4x-ultrasharp.pth | 0 options: |
| iterations | INT | 31–10000 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| hires_cfg | SP_HiresGen_HiresCfg | — |