High Resolution v2

This is Searge's take on the hi-res fix - the two-pass trick where you generate at SDXL's native resolution, then run a second pass at a larger size with low denoising to add coherent detail instead of just more pixels. SeargeHighResolution is the control panel for that pass in the v4 workflow, and it bundles in extra dials for detail, contrast, saturation, and a separate final upscale on top.

It's the pack's improved replacement for the old "Hi-Res Fix," and it lives under Searge/UI/Inputs, so it's current v4.x. Searge specifically called out better high-resolution modes as a v4 improvement over the earlier workflow's hires handling.

How it works

The core is a diffusion upscale: the image goes up in size and gets a light denoising pass so the model paints in detail that's consistent with the larger canvas, rather than the soft, stretched look you get from plain interpolation. The key knob is hires_denoise - and its default of 0.2 is telling. Hi-res fix wants low denoise; go above ~0.5 and the composition starts changing instead of just gaining detail, which is the single most common way people wreck this pass. Keeping it in the 0.2–0.4 band preserves the original while sharpening it.

On top of that, the node exposes tone and detail adjustments (softness, detail boost, contrast, saturation) and a latent_detailer option, plus a final_upscale_size that applies a further enlargement after the hi-res pass - handy for going from a hi-res 1.5× straight up to a 2–4× final without a separate node.

The inputs and outputs that matter

You won't touch all of these; the ones that count:

• hires_mode - none, simple - fast, or normal. Off, quick, or full.
• hires_scale - 1.25x, 1.5x (default), or 2x. How much the hi-res pass enlarges.
• hires_denoise (default 0.2) - the make-or-break dial. Keep it low.
• final_upscale_size - an optional extra upscale (1.2x up to 4.0x) after the hi-res pass.

The rest - hires_softness, hires_detail_boost, hires_contrast_factor, hires_saturation_factor, hires_latent_detailer - are finishing tweaks; leave them at defaults until you have a specific look in mind. Connectors are the data in/out (SRG_DATA_STREAM).

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: python -m pip install opencv-python in ComfyUI's Python env (required), then cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, restart. Windows portable has an installer script.

The final-upscale path uses an ESRGAN model, so have one (e.g. 4x-UltraSharp) in ComfyUI/models/upscale_models. The hi-res diffusion pass itself uses your loaded SDXL checkpoint.

Where people get burned

• Denoise too high. Above ~0.5 the hi-res pass rewrites your image instead of refining it. If your careful composition drifts during upscaling, this dial is why - pull it back to 0.2–0.4.
• Stacking huge scales. A 2× hi-res plus a 4× final is a lot of pixels and a lot of VRAM; the diffusion pass is the slow, memory-hungry part. Scale in sensible steps.
• Over-boosting the finishing sliders. Detail boost, contrast, and saturation are easy to overdo. A little goes a long way; cranking them gives you a crunchy, oversaturated result.