<h1 id="sexiam's-sdxl-txt2img-workflow-(comfyui)">🔥 Sexiam’s SDXL Txt2Img Workflow (ComfyUI)</h1>A simple, reliable workflow built for clean detail, strong structure, and automatic Civitai metadata detection.If you want a workflow that “just works” — clean generations, sharp details, and all your LoRAs + prompts properly detected by Civitai — this walkthrough covers exactly how my setup runs.V2 of the workflow added the ability to turn off the upscale by model group. In v1 the image wouldn't save. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/aab2f91e-52c8-4e09-948e-6e8102ad471e/width=525/aab2f91e-52c8-4e09-948e-6e8102ad471e.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c2177b62-872a-4087-9b56-a5c1dbf1aa51/width=525/c2177b62-872a-4087-9b56-a5c1dbf1aa51.jpeg" /><hr /><h1 id="1-load-your-checkpoint-(and-vae-if-you-want)">1️⃣ Load Your Checkpoint (and VAE if you want)</h1>Start by loading your SDXL checkpoint. If you prefer a different VAE, load it here — the toggle swaps between the checkpoint’s VAE and whatever VAE you manually provide.This gives you full control over color, clarity, and contrast right at the start.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0a27ab14-d524-4606-a5b6-c783c28230fa/width=525/0a27ab14-d524-4606-a5b6-c783c28230fa.jpeg" /><hr /><h1 id="2-add-your-loras">2️⃣ Add Your LoRAs</h1>Hit Add LoRA → pick what you want.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/398dbac2-aaa9-41da-a2b1-73a8eb8d5f07/width=525/398dbac2-aaa9-41da-a2b1-73a8eb8d5f07.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/57edbc66-43a9-421f-8365-83fdb5d32ace/width=525/57edbc66-43a9-421f-8365-83fdb5d32ace.jpeg" />Pro move: Right-click a LoRA → ComfyUI instantly shows its trigger words. Copy them into your prompt and you’re already ahead of 90% of users.<hr /><h1 id="3-pick-your-image-size-(latent-space)">3️⃣ Pick Your Image Size (Latent Space)</h1><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e8c24965-37f0-425e-9d77-3cf681f5bf21/width=525/e8c24965-37f0-425e-9d77-3cf681f5bf21.jpeg" />Use the Latent Size Picker to select your standard SDXL 1MP resolutions.Want more detail in the base generation?<ul><li>Add Upscale Latent By</li><li>Set the value to 1.5×</li></ul>1.5× is the sweet spot. Higher than that = more errors, weird proportions, or flat results since SDXL isn’t trained on massive latent sizes.<hr /><h1 id="4-write-your-prompts-(positive-+-negative)">4️⃣ Write Your Prompts (Positive + Negative)</h1><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8b15fbc2-19c4-4ccd-b6b4-219cb414e074/width=525/8b15fbc2-19c4-4ccd-b6b4-219cb414e074.jpeg" />Fill out your positive and negative prompts using the green and red TextEncode nodes.If you’re planning to post on Civitai:<h3 id="use-impactwildcardencode">👉 Use ImpactWildcardEncode</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b72db708-51cd-40d0-ac74-1cd8dd2e7982/width=525/b72db708-51cd-40d0-ac74-1cd8dd2e7982.jpeg" />It automatically embeds LoRA filenames into your metadata so Civitai recognizes everything — model, LoRAs, sampler, steps, seed, prompts, the whole stack.No manual cleanup needed.<hr /><h1 id="5-two-pass-ksampler-system-(the-core-of-the-workflow)">5️⃣ Two-Pass KSampler System (The Core of the Workflow)</h1>This is the heart of the setup: a clean, controlled generation pass followed by a refinement pass.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/465634dc-d9e0-4a6d-8d56-fb71181ade1b/width=525/465634dc-d9e0-4a6d-8d56-fb71181ade1b.jpeg" /><h3 id="ksampler-1-full-generation">KSampler #1 — Full Generation</h3><ul><li>Denoise: 1.00 This creates your full image from noise. Think of it as the “big picture” pass.</li></ul><h3 id="ksampler-2-refinement">KSampler #2 — Refinement</h3><ul><li>Denoise: 0.45 This tightens edges, improves materials, sharpens lighting, and adds definition without overwriting the original structure.</li></ul>Together, they give you consistency + clarity without overbaking your images.<hr /><h1 id="6-optional:-third-pass-+-model-upscaling">6️⃣ Optional: Third Pass + Model Upscaling</h1><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8f73c2cf-d195-46bc-aeb9-bb20daf93615/width=525/8f73c2cf-d195-46bc-aeb9-bb20daf93615.jpeg" />Want a print-ready or ultra-crisp final?This workflow includes a third KSampler connected to a full Upscale By Model system:<ol><li>Third sampler reprocesses the refined latent</li><li>Image Upscale With Model applies your favorite upscale model (4xRemacri, UltraSharp, etc.)</li><li>Upscale Image % sets your final dimension boost</li></ol>You get clean high-res results without blowing up faces, hands, or fine textures.<hr /><h1 id="7-final-output-(civitai-compatible-metadata)">7️⃣ Final Output (Civitai-Compatible Metadata)</h1><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/47580c9f-38bf-4777-8c5e-ba1877005d01/width=525/47580c9f-38bf-4777-8c5e-ba1877005d01.jpeg" />The last node handles everything automatically.It records:<ul><li>Checkpoint name</li><li>LoRAs used</li><li>Positive prompt</li><li>Negative prompt</li><li>Steps</li><li>CFG</li><li>Sampler + Scheduler</li><li>Width / Height</li><li>Seed</li><li>Upscale model</li><li>And more</li></ul>Upload to Civitai and it all shows up perfectly — no missing tags, no guessing, no manual editing.

Sexiam - Txt2Img with Upscaler

## 🔧 What Changed in V21. Metadata Save Node Was RebuiltIn the original version, the Metadata Save / Image Saver node could not function properly when the Upscale-By-Model chain was turned off. In V2, I replaced that with a newer <code>comfyui-image-saver</code> node and rebuilt the input links:<ul><li>It now accepts all required parameters, even when the upscale branch is bypassed.</li><li>It also includes the updated widget values such as filename formatting, date stamp, and metadata toggles.</li></ul>This ensures it always receives a valid image input, so metadata is saved even when you disable the upscale.## 🎯 Why These Changes Were Needed✔ Metadata Saving Was FailingYour previous saver node only worked if the upscale path was active, because it expected a connected image output. V2 fixes that by ensuring the saver always receives:<ul><li>an image</li><li>all metadata fields</li><li>prompts</li><li>sampler settings</li></ul>This means uploads to Civitai always include:<ul><li>model</li><li>LoRAs</li><li>sampler info</li><li>width/height</li><li>seed</li><li>CFG</li><li>positive/negative prompts</li><li>hashes</li><li>A1111 metadata format</li></ul>All without requiring the upscale.✔ Upscaling Is Now Truly OptionalBefore V2:<ul><li>Disabling the upscale = broken workflow.</li></ul>After V2:<ul><li>Upscaling on → works</li><li>Upscaling off → also works</li><li>Metadata → always saved properly</li></ul>The bypass node makes the whole section modular.

Files with SHA256: c711501c2152049f02e27bdbeb53a266fe81faf647801964fa3e33a02b3eb3bb