<hr /><h1 id="caption-+-sdxl-adetailer-+-upscale-workflow-overview">📌 Caption + SDXL aDetailer + Upscale — Workflow Overview</h1>This workflow is designed as an end-to-end enhancement pipeline for improving existing images using three major stages:<ol><li>Automatic Captioning (JoyCaption)</li><li>Multi-stage SDXL aDetailer refinement</li><li>High-quality 2×/4× upscaling</li></ol>It is especially useful for batch-processing portrait images, NSFW or SFW, while keeping full control over detail restoration (faces, hands, skin, body parts), composition integrity, and final resolution.<hr /><h2 id="1.-automatic-captioning-(joycaption)">🔹 1. Automatic Captioning (JoyCaption)</h2>The pipeline begins with JoyCaption:<ul><li>Loads images from a directory via ImageBatchPath</li><li>Applies the user-defined Extra Options block (lighting, camera info, text removal, etc.)</li><li>Produces a clean SDXL-style prompt</li><li>The prompt is shown via ShowText and sent directly into the conditioning stack</li></ul>This step is ideal for:<ul><li>Captioning large datasets</li><li>Creating prompts from real photos</li><li>Generating consistent SDXL prompts for further enhancement or LoRA training</li></ul><hr /><h2 id="2.-multi-stage-sdxl-adetailer-refinement">🔹 2. Multi-Stage SDXL aDetailer Refinement</h2>The core of the workflow is a chain of aDetailer passes, each dedicated to a specific target area:<ul><li>Face</li><li>Hands</li><li>Teeth</li><li>Armpits</li><li>Breasts</li><li>Vagina</li><li>General body segmentation</li></ul>Each pass includes:<ul><li>YOLOv8 detectors from the Impact Pack</li><li>SAM for accurate mask generation</li><li>SDXL inpaint refinement with positive/negative conditioning</li><li>Full previews for each refinement stage</li></ul>This allows extremely fine-grained control and produces:<ul><li>Cleaner face geometry</li><li>Corrected hands</li><li>Better skin continuity</li><li>Consistent anatomy in NSFW work</li><li>Reduced artifacts from low-quality images</li></ul>The workflow is modular: users can disable or remove specific detailers as needed.<hr /><h2 id="3.-high-resolution-upscaling-(2-or-4)">🔹 3. High-Resolution Upscaling (2× or 4×)</h2>Upscaling is handled through:<ul><li>UpscaleModelLoader (4x_NMKD-Siax_200k)</li><li>ImageUpscaleWithModel for AI-based enhancement</li><li>ImageScaleBy for additional clean integer scaling</li><li>Two final outputs:<ul><li>2× upscaled</li><li>4× upscaled</li></ul></li></ul>This results in:<ul><li>Clean, crisp edges</li><li>Preservation of style</li><li>Better texture definition</li><li>High-resolution images suitable for posting or further processing</li></ul>The Set/Get image node pair provides a lightweight way to reuse the intermediate image cleanly across the upscale branch.

Files with SHA256: 48d4905bfacc264c07ddd1530c9b4568eeeaee003158ccb82e49c94c6a27439a