<h3 id="i've-tried-to-create-the-easiest-ultimate-sd-upscale-workflow-possible.">I've tried to create the easiest Ultimate SD Upscale workflow possible.</h3>SFW/NSFW, doesn't matter.This workflow is intended for upscaling the already-awesome images you've created.I looked around CivitAI for a workflow like this without spaghetti noodles. I guess I have a way of making workflows where all the nodes are close together in a box, kinda like an Automatic1111 or SwarmUI interface. Because once everything is configured correctly, why should you have to move around an enormous canvas? Anyway...<hr />This workflow assumes you have a general understanding of what you're doing. That includes things like:<ul><li>Why you're using SD Ultimate Upscale in the first place</li><li>What upscale models you like to use</li><li>The checkpoint/LoRAs/sampling parameters you want to use<ul><li>I usually use identical model/LoRA/cfg/sampler/scheduler as my original image with a 0.30 denoise. Positive prompt is just quality tags for me, like '8k, absurdres, masterpiece', but this depends on what you're aiming for.</li></ul></li><li>All of the parameters in the Ultimate SD Upscale node below the sampler/scheduler stuff( mask_blur, tile_padding, etc etc), are intended to be modified to get your desired output. What I have in the workflow is what works for my images but YOU need to tweak it and iterate.</li></ul><hr /><h3 id="what-models-does-it-work-with">What models does it work with?</h3> This workflow is intended for any SDXL-based model that uses the standard checkpoint file you find on CivitAI. I use Illustrious exclusively but I don't see why it shouldn't completely work 100% with Pony, NoobAI, etc.<hr /><h2 id="custom-nodes">Custom nodes</h2>As always, I try to use core ComfyUI nodes as much as possible to reduce bloat and annoyance. This workflow requires 4 custom node packs to work:ComfyUI-Custom-Scripts (for the math expression nodes in the Backend subgraph):<a target="_blank" rel="ugc" href="https://github.com/pythongosssss/ComfyUI-Custom-Scripts">https://github.com/pythongosssss/ComfyUI-Custom-Scripts</a>rgthree-comfy (for the awesome LoRA loader and image comparer):<a target="_blank" rel="ugc" href="https://github.com/rgthree/rgthree-comfy">https://github.com/rgthree/rgthree-comfy</a>ComfyUI_UltimateSDUpscale (for the upscaling node):<a target="_blank" rel="ugc" href="https://github.com/ssitu/ComfyUI_UltimateSDUpscale">https://github.com/ssitu/ComfyUI_UltimateSDUpscale</a>ComfyUI_tinyterraNodes (for the positive and negative prompt nodes):<a target="_blank" rel="ugc" href="https://github.com/TinyTerra/ComfyUI_tinyterraNodes">https://github.com/TinyTerra/ComfyUI_tinyterraNodes</a><hr /><h3 id="how-to-use-it">How to use it</h3><ol><li>Drag any image onto “Load Input Image” (or choose a file).</li><li>Add LoRAs as needed.</li><li>Choose your upscale model (e.g., Real-ESRGAN, NMKD Siax, UltraSharp, etc.)</li><li>Change “Upscale By…” to 2 (default = 2×), 4 (4×), 1.5, 3, whatever you want. This node increments in steps of 0.5.</li><li>Add your positive and negative prompts.</li><li>In the Ultimate SD Upscale node, change around the steps/cfg/sampler/scheduler/denoise as you wish.</li><li>Hit Queue Prompt → Done.</li></ol>What you get instantly:<ul><li>An upscaled image that respects your original aspect ratio exactly.</li><li>Saves the final upscaled image straight to your output folder.</li><li>Built-in side-by-side slider (rgthree Image Comparer) so you can see before/after instantly.</li></ul><hr /><h3 id="what-happens-behind-the-scenes">What happens behind the scenes</h3>There’s one tiny orange “Backend” box that does all the magic automatically:<ul><li>Detects your input resolution</li><li>Multiplies width &amp; height by your upscale factor → calculates exact target dimensions</li><li>Feeds those target dimensions into proper SDXL-style conditioning (positive + negative)</li><li>Gives the tiling sampler perfect knowledge of the final size it’s aiming for</li></ul>Feel free to open or expand the subgraph as needed.

Files with SHA256: a4c94caf1d754add9160c3c4017a89b7838cc1da04da75545cfbeafb342441f6