bing-su/adetailer

ADetailer is an extension for the AUTOMATIC1111 Stable Diffusion web interface, the popular browser-based tool people use to generate images with AI. The extension adds an automatic step that detects specific objects in a generated image and then re-draws (inpaints) those areas with extra attention, without requiring the user to do any manual masking. The main problem it solves is a common quality issue with AI image generation: faces, hands, and other detailed body parts often come out distorted or inconsistent, especially in images where they are small or appear off to the side. ADetailer runs detection on the finished image, finds those areas automatically, and runs an additional inpainting pass on just those regions to sharpen them up. The detection side uses YOLO models, which are object-detection models trained to find specific targets in an image. ADetailer ships with pre-trained models for faces (both realistic and illustrated styles), hands, and full person segments. You pick the model that matches what you want to fix, and the extension handles the rest. You can also add your own custom YOLO models by dropping them into a specific folder. The inpainting step can use the same prompts and settings as your original generation, or you can give ADetailer its own separate prompt to guide how the touched-up area should look. There are controls for confidence thresholds, mask size filters, mask adjustments like erosion and dilation, and integration with the ControlNet extension if you have that installed. Installation is done directly inside the Extensions tab of the Stable Diffusion web UI, either from the built-in extension browser or by pasting in the repository URL. No separate download or command-line steps are needed beyond restarting the interface after install.