This Generation Guide will discuss the way I recommend using two of the custom checkpoint merges I have uploaded together to achieve maximum results.
Warlock - This is the architect. The model excels at prompt adherence/scene composition and is especially effective when adding Danbooru tags to your prompts.
Piledriver - This is the closer. The model compliments Warlock by finishing the images in a psuedo-realistic style.
Workflow:
VAE - Liquid9745 or SDXL Natural Skintone (FP32 VAE)
Sampling method - DPM++ 2M
Schedule type - Karras
Sampling steps - 40
CFG Scale - 7
Checkpoint - Warlock
Refiner - Piledriver
Switch at - 0.4
Prompt - realistic, photorealistic
These settings will give you a strong start on your SDXL renders, tweak them to taste. Below I will expand on the use of ADetailer
I change the following settijngs -
Mask Preprocessing - Mask erosion (-) / dilation (+) - 16
Inpainting - Inpaint denoising strength - 0.48
Inpainting - (✓)Use separate VAE - Liquid9745
To help automate and maximize the use of ADetailer, use the embedding detfac (Detailed Face Embedding) in the positive prompt box inside the extensions UI (user interfece). The embedding is gender/race/ect neutral, it is simply compact prompt instructions for ADetailer to render a detailed, high quality face.
You can prompt for multiple faces in a single image. Render with ADetailer deactivated for faster compute times. When you achieve a render you like but wish to improve the faces, recycle the seed, activate ADetailer, make sure you've set a detector model such as face_yolov8n.pt,
adjust the settings to taste, (if using WebUI) make sure in the Settings tab that the option for Sort bounding boxes by is set to Position (left to right), write your ADetailer prompt for the face furthest to the left in the image, then add this syntax - "[SEP]" (without paranthesis), now write your prompt for the next face to the right of the first face, ect, ect.
### Automatic1111 Stable Diffusion WebUI
1. Launch the WebUI in your browser.
2. Navigate to the "Extensions" tab.
3. Click "Available" and then "Load from" to fetch the extension list.
4. Search for "adetailer" in the search bar.
5. Click "Install" next to the ADetailer extension by Bing-su.
6. Restart the WebUI or click "Apply and restart UI" if prompted.
7. Download detection models (e.g., face_yolov8n.pt) from the extension's GitHub and place them in the models/adetailer folder.
### ComfyUI
ADetailer is not natively available as an extension, but equivalent face/hand detailing can be achieved using custom nodes like Face Detailer from the ComfyUI-Impact-Pack.
1. Install ComfyUI Manager if not already (via git clone in custom_nodes folder).
2. In ComfyUI, open the Manager menu.
3. Search for "ComfyUI-Impact-Pack" and install it.
4. Restart ComfyUI.
5. Use nodes like "FaceDetailer" or build a workflow with detection models (e.g., YOLO) for auto-masking and inpainting similar to ADetailer.
### Fooocus
ADetailer is not directly supported, but Fooocus v2.5+ includes a built-in "Enhance" feature that provides similar auto-detection and refinement for faces/hands without needing installation.
1. Update to Fooocus v2.5 or later (via git pull or download from GitHub).
2. In the interface, enable "Enhance" in the advanced settings for automatic detailing during generation.
### InvokeAI
ADetailer is not available as a built-in or extension. A feature request exists for similar functionality, but currently, you can approximate it manually via workflows using masking and inpainting nodes.
1. No direct install; use InvokeAI's graph editor to create a custom workflow with face detection (requires external models/scripts).