Advanced Parameters

Advanced Parameters is the node where you control the heart of SDXL's two-model setup in the Searge workflow: how much work the base model does versus the refiner, how strong the refiner pass is, and a few batch and upscale settings. If you want to understand what makes a Searge image, this is one of the nodes to read.

It's part of SeargeSDXL, the SDXL base-plus-refiner pack Searge released at SDXL 1.0's launch in 2023 - one of the first workflows to wire the base and refiner checkpoints together cleanly. This node lives in Searge/_deprecated_/UI/Inputs; the v4.x rewrite replaced the input-chain design with a data stream, but the parameters here are the same ones any SDXL two-stage setup exposes.

How it works

SDXL 1.0 was designed as a two-stage system: a base model does most of the denoising, then a refiner model finishes the last chunk of steps to add fine detail. The base_ratio is exactly that split - 0.8 means 80% of the steps run on the base and the last 20% hand off to the refiner. refiner_strength and refiner_intensity (soft/hard) then tune how aggressively that refiner pass acts.

This node collects those settings and adds them to the PARAMETER_INPUTS bundle flowing through the input chain - it doesn't sample anything itself. The actual generation reads these values back out downstream.

The inputs that matter

• base_ratio (FLOAT, default 0.8) - the base/refiner step split. The single most consequential knob here.
• refiner_strength (FLOAT, default 1) and refiner_intensity (soft/hard) - how hard the refiner pass hits.
• denoise (optional FLOAT, default 1) - full denoise for txt2img; lower it for img2img so you keep the source.
• batch_size (INT, 1–4) and precondition_steps (INT, 0–10) - batch count and a few warmup noise steps.
• upscale_resolution_factor (FLOAT, default 2) and save_upscaled_image - the final upscale multiplier and whether to keep it.
• inputs in / inputs out (PARAMETER_INPUTS) - the chain wire.

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 require python -m pip install opencv-python in ComfyUI's Python environment first, or the README warns the pack won't load. You'll also want the models: SDXL base and refiner checkpoints (the refiner is what half this node controls), the fp16-fix VAE, and upscale models for the upscale factor. The README has direct links.

Common issues

Here's the honest one: the refiner these settings exist to control is a feature the community has largely abandoned. The realistic-checkpoint crowd summed it up as "it always seemed like an added step that never added much," and most people now run a single fine-tuned SDXL checkpoint at base_ratio effectively 1.0 - no refiner at all. If your refiner is making images worse (over-smoothed, plasticky), try pushing base_ratio toward 1.0 or dropping refiner_strength.

For img2img, remember to lower denoise - leaving it at 1 throws your source image away entirely. And as always with Searge: latest workflow JSON, latest nodes, or version drift will bite you.