Generation Parameters

SeargeInput2 is the single node that used to hold most of the dials for a Searge SDXL run - seed, size, steps, CFG, sampler, scheduler, and where to save the output - all in one box. It doesn't generate anything itself. It gathers your generation settings and passes them, as one bundle, into the rest of the workflow.

Two things to know up front. First, this is an older-generation node: its category is Searge/_deprecated_/UI/Inputs, which means it belongs to the v2/v3-era workflow. The current v4.x pack replaced it with SeargeGenerationParameters (labelled "v2"), which does the same job with SDXL size presets and a cleaner data stream. Second, it's a Searge node, so it only makes sense inside Searge's own graph - it's not a general settings panel you can bolt onto any workflow.

How it works

Everything on this node is a widget you set by hand, and the node's only output is the whole collection, packed into Searge's PARAMETER_INPUTS type. Downstream nodes unpack that bundle and use the pieces - the sampler node reads steps/CFG/sampler/scheduler, the latent setup reads width/height, the save node reads the save toggle. Bundling settings this way is the whole idea behind Searge's design: keep the tweakable knobs in one place so you're not chasing them across a dozen scattered nodes.

There's also an optional inputs passthrough, so you can chain another parameter source in and let this node extend it rather than replace it.

The inputs and outputs that matter

You could touch all of these, but the ones that actually change your image:

• seed - the usual randomness anchor. Fix it to reproduce a result.
• image_width / image_height - both default to 1024, the native SDXL resolution. Stay on SDXL's trained aspect ratios (1024×1024, 1152×896, 1216×832, and their rotations); off-ratio sizes stretch anatomy.
• steps (default 20) and cfg (default 7) - quality-vs-time and prompt-adherence-vs-freedom.
• sampler_name - 44 options. The default here is ddim, which is a bit of a period piece; for SDXL the community workhorse is DPM++ 2M Karras. That's a DDPM-style model, so the classic reshaped-schedule advice genuinely applies to it.
• scheduler - defaults to ddim_uniform; karras is the pairing most people actually want with SDXL.
• save_image and save_directory - whether to write the result and to which folder.

Output is the single inputs bundle (PARAMETER_INPUTS).

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: python -m pip install opencv-python in ComfyUI's Python env first (it's a real dependency, not optional), then cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, restart. Windows portable users get a one-click installer script that can also fetch the SDXL base/refiner checkpoints and upscalers the workflow expects.

Where people get burned

• The ddim defaults. They're fine, but they're not what most SDXL guides assume. If your output looks soft or dated, swap to DPM++ 2M Karras with the karras scheduler before blaming anything else.
• Using it in a new build. It's deprecated. For anything current, load the v4.x workflow and use SeargeGenerationParameters instead - you get SDXL aspect-ratio presets and the base-vs-refiner ratio in the bargain.
• Off-ratio dimensions. SDXL was multi-aspect trained but not resolution-agnostic. Generate at a trained ratio and upscale after; don't type 1920×1080 in here and expect clean anatomy.