HiResFix Parameters

HiResFix Parameters (the output node) unpacks the high-resolution-pass settings from the Searge parameters bundle - the second-pass denoise, upscale factor, steps, and seed - into individual wires the workflow uses to run its hi-res fix. It's the read-back partner to the HiResFix Parameters input node.

It's from SeargeSDXL, the SDXL base-and-refiner workflow pack Searge released at SDXL 1.0's launch in 2023. It lives in Searge/_deprecated_/UI/Outputs - v3-era internal wiring the v4.x data-stream rewrite replaced - and you don't place it by hand; it's part of the workflow.

How it works

The Searge workflow bundles all settings into one PARAMETERS object and fans pieces back out with output nodes. This one carries the hi-res-fix slice. Hi-res fix is the two-pass trick: generate at SDXL's native ~1024px, upscale the latent, then run a second sampling pass at low denoise to add real detail instead of stretching pixels. The knowledge base calls it "essential for any generation above native resolution." This node surfaces the values that control that second pass, and passes the bundle through for chaining.

The outputs that matter

• hires_fix (ENABLE_STATE) - whether the hi-res pass runs at all.
• hrf_denoise (FLOAT) - how much the second pass may change the image. The critical one: keep it low (~0.3–0.5) so you add detail without altering the composition.
• hrf_upscale_factor (FLOAT) - how much bigger; 1.5–2x is the sane range.
• hrf_steps (INT) - steps for the pass.
• hrf_seed (INT) and hrf_noise_offset (INT) - the second-pass seed and offset.
• hrf_smoothness (FLOAT) - smoothing on the pass.
• parameters (PARAMETERS) - chain-through.

Nothing is set here; the values come from the input-side node.

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, then restart. Manual installs need python -m pip install opencv-python in ComfyUI's Python environment first, per the README. Hi-res fix works best with a good upscale model like 4x-UltraSharp - the README lists it alongside the SDXL base/refiner checkpoints.

Common issues

The recurring hi-res mistake shows up through the value this node carries: hrf_denoise too high. Above ~0.5 the knowledge base warns "the composition changes" - you get a different image, not a bigger one. If the result doesn't match the original, that denoise is set too high in the input node (this node only reports it). If nothing changes at all, it's too low.

Also relevant: in early v4.0 the hi-res latent detailer was mis-wired and did nothing; the pack fixed it in v4.1 - another reason to run the latest node version against the latest workflow JSON. And the honest framing: this is deprecated plumbing from a refiner-era workflow. Hi-res fix remains a great technique, but you can get it in any modern SDXL workflow without Searge.