Finally sharing my stable, production-ready head swap workflow for ComfyUI. No custom LoRA training needed.What it does:<ul><li>Swaps heads from source → target image while preserving pose, lighting, and scene context</li><li>Runs on Qwen Edit + Lightning LoRA (4 steps) – super fast</li><li>Automatic face masking, cropping, and intelligent blending</li><li>Smart system prompts generated from both images via vision models</li><li>ControlNet + depth/structure for alignment control</li><li>Final upscale with SeedVR2 for clean, polished output</li></ul>Key features: ✅ Fast (4-step Lightning LoRA) ✅ Automatic mask generation &amp; refinement ✅ Pose + expression + lighting preservation ✅ High-resolution upscaled outputWorkflow includes:<ul><li>AutoCropFaces for source head detection</li><li>FaceSegment for automatic face masking</li><li>Gemini vision analysis for auto-generated system prompts</li><li>ControlNet with DepthAnythingV2 + DW Preprocessor</li><li>InpaintModelConditioning to keep backgrounds untouched</li><li>SeedVR2 + blend for final upscale</li><li>Resolution auto-management to protect your VRAM</li></ul>Best results for:<ul><li>Portrait photography</li><li>Half-body shots</li><li>Consistent lighting scenes</li><li>When you want natural-looking identity swaps</li></ul>Customization tips:<ul><li>Lower ControlNet weights if you want looser/weirder edits</li><li>Increase MaskGrow blur for softer blends</li><li>Add custom prompt for some extra-details.</li></ul><a target="_blank" rel="ugc" href="https://huggingface.co/olesheva/head_swap_qwen_edit">https://huggingface.co/olesheva/head_swap_qwen_edit</a>Support my work &lt;3<a target="_blank" rel="ugc" href="http://Ko-fi.com/oleshevakatya">Ko-fi.com/oleshevakatya</a>

Files with SHA256: 4936f594876c2b3be92858332de936d10dbeaeea2770920c2acfd4e324462ff7