<h1 id="sexiam's-img2img-+-upscale-workflow-guide-(comfyui)">Sexiam’s IMG2IMG + Upscale Workflow Guide (ComfyUI)</h1>A clean, practical walkthrough for using the workflow effectively.<hr /><h1 id="1.-load-your-checkpoint">🧱 1. Load Your Checkpoint</h1>Start here:<ul><li>Load your SDXL checkpoint</li><li>(Optional) Load an external VAE</li></ul>SDXL behaves differently depending on its VAE, so pick whichever looks best for your model.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ddf546bd-8cad-4749-bf59-7ddbf7299b4c/width=525/ddf546bd-8cad-4749-bf59-7ddbf7299b4c.jpeg" /><hr /><h1 id="2.-set-your-vae-mode">🔧 2. Set Your VAE Mode</h1>If you want to use the checkpoint’s built-in VAE, set:<ul><li>“Use Checkpoint VAE?” = True</li></ul>If you want to use your external VAE, set:<ul><li>“Use Checkpoint VAE?” = False</li></ul><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/af114f75-1383-48a8-a3fc-39500b1405ef/width=525/af114f75-1383-48a8-a3fc-39500b1405ef.jpeg" /><hr /><h1 id="3.-load-your-input-image">🖼️ 3. Load Your Input Image</h1>Upload your source image using the Load Image node.<h3 id="orientation-check">Orientation Check</h3>Set the “Image is Portrait?” toggle correctly:<ul><li>Portrait → True</li><li>Landscape → False</li></ul><h3 id="load-an-upscale-model">Load an Upscale Model</h3>Pick the upscaler you want (Remacri, SwinIR, ESRGAN, etc.). You’re not upscaling the final output here — you’re shaping a clean latent size for SDXL.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/afc2d1a9-405f-4ba8-a973-03ab88a53b46/width=525/afc2d1a9-405f-4ba8-a973-03ab88a53b46.jpeg" /><hr /><h1 id="why-the-workflow-uses-'upscale-by-model'-before-sampling">📏 Why the Workflow Uses ‘Upscale by Model’ Before Sampling</h1>You might ask: “Why am I using ‘Upscale by Model’ for image-to-image?”Here’s the short answer:➡️ You’re not actually upscaling to output — you’re resizing the latent. SDXL was trained on ~1MP images, so pushing too high causes:<ul><li>Generation errors</li><li>Warped structure</li><li>Model collapse</li></ul>This workflow uses a safe baseline:<ul><li>832×1216 (portrait)</li><li>1216×832 (landscape)</li><li>Multiplied by 1.5× (50% larger)</li></ul>This is the maximum size SDXL samplers can reliably handle before breaking.If you're having issues, reset the “Scale to Target Ratio” nodes to whatever size works for your system.This workflow is ideal for using the input as a loose generational base, meaning:<ul><li>Great for creative reinterpretations</li><li>Fine for refinement</li><li>Some proportions/details may change (normal for IMG2IMG)</li></ul><hr /><h1 id="4.-enter-your-prompts">✍️ 4. Enter Your Prompts</h1>Fill in your:<ul><li>Positive prompt</li><li>Negative prompt</li></ul>These control the details, style, and adherence to the original image.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/62b77b11-ec77-4071-914b-ebb92fff39e1/width=525/62b77b11-ec77-4071-914b-ebb92fff39e1.jpeg" /><hr /><h1 id="5.-adjust-ksampler-settings-(critical-for-img2img)">🎛️ 5. Adjust KSampler Settings (Critical for IMG2IMG)</h1>For IMG2IMG, denoise strength is the most important setting:<ul><li>0.7 → Loose interpretation of the input<ul><li>Model only reuses ~30% of the original</li></ul></li><li>0.5 → Preserves composition + color<ul><li>Allows detail refinement</li><li>Great for keeping the structure mostly intact</li></ul></li></ul>Use lower values when you want accuracy, higher when you want creativity.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a974ce01-65ba-4880-8f10-6e03c9f11e40/width=525/a974ce01-65ba-4880-8f10-6e03c9f11e40.jpeg" /><hr /><h1 id="6.-optional:-final-model-upscaling-(use-only-at-low-denoise)">🔍 6. Optional: Final Model Upscaling (Use Only at Low Denoise)</h1>This section lets you upscale the final output if your denoise is 0.4 or lower.Why this matters: Low denoise keeps most of the latent structure, so upscaling is stable.Results vary based on:<ul><li>GPU VRAM</li><li>System RAM</li><li>How large you upscale</li></ul><h3 id="steps:">Steps:</h3><ul><li>Load the final image</li><li>Set the upscale multiplier<ul><li>1.5× → 50% bigger</li><li>2× → double the resolution</li></ul></li><li>Choose an upscale model<ul><li>Realistic outputs → clean general ESRGAN models</li><li>Stylized/anime outputs → anime-optimized upscale models</li></ul></li></ul><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/32ff553e-db2e-425b-ae9c-37707ffa4be2/width=525/32ff553e-db2e-425b-ae9c-37707ffa4be2.jpeg" /><hr /><h1 id="required-custom-nodes"># 📦 Required Custom Nodes</h1>These are the only node packs used in the workflow. Install them through ComfyUI-Manager or manually via GitHub.<a rel="ugc" href="https://github.com/rgthree/rgthree-comfy">rgthree-comfy</a><a rel="ugc" href="https://github.com/MixLabPro/comfyui-mixlab-nodes">comfyui-mixlab-nodes</a><a rel="ugc" href="https://github.com/MixLabPro/comfyui-mixlab-nodes">Masquerade Nodes</a><a rel="ugc" href="https://github.com/M1kep/ComfyLiterals">ComfyLiterals</a>

Files with SHA256: 0fb689f3bf8719fd60b6203301383addf3b976db54c40c5d4ac379bfc56ce147