This model will help you remove unwanted artifacts and imperfections from faces in your SDXL image generations, resulting in cleaner and more polished results.

negdetfac - Negative Embedding for Detailed Face (SDXL)

Model Type: Textual Inversion (Embedding) Compatible With: all SDXL models (Pony, Illustrious, etc.) Version: 1.0 File: negdetfac.safetensors Trigger Words: negdetface

Description

This embedding is a negatives-only synthetic textual inversion embedding created through runtime token merging. It is designed to complement the positive "detfac" embedding by removing common unwanted facial elements and artifacts in SDXL generations. It targets issues like deformed or asymmetrical features, blurry or mutated eyes, bad skin textures, extra or missing facial parts, unnatural distortions, and other common face-related flaws. When used alongside detfac, it helps achieve balanced, high-quality photorealistic faces by enhancing details positively while actively suppressing defects—ideal for portraits, character close-ups, or any scene requiring clean human faces.

Usage

• In Main Negative Prompt: Add "negdetface" to your negative prompt for global cleanup of facial artifacts, e.g.: Positive prompt: "photorealistic portrait of a woman, detfac, masterpiece, 8k" / Negative prompt: "negdetfac". Adjust strength with weights like (negdetfac:1.1) for stronger removal of flaws.

• With ADetailer Face Models: Place "negdetfac" in ADetailer's negative prompt box during auto-inpainting (e.g., with face_yolov8s.pt model). This provides targeted artifact removal on detected faces, pairing perfectly with detfac in the positive box for refined, defect-free results without manual prompt tweaking—great for batch processing or complex compositions.

Examples

• Base Prompt Without: Often includes deformed eyes, asymmetrical features, bad anatomy, or skin artifacts.

• With negdetfac (especially alongside detfac): Eliminates unwanted flaws for cleaner, symmetrical, and naturally detailed faces.