A simple workflow to help you get high resolution images with deep details quickly.<a target="_blank" rel="ugc" href="https://ku-cvlab.github.io/Perturbed-Attention-Guidance/">Perturbed-Attention Guidance (PAG)</a> and <a target="_blank" rel="ugc" href="https://github.com/Extraltodeus/ComfyUI-AutomaticCFG">Automatic CFG</a> nodes help bring additional detail and quality into image generations with little extra work by the user.<a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/cosxl">COSXL</a> is a new experimental checkpoint model from StabilityAI that can produce images with wider range of color, from 'pitch black' to 'pure white'.This workflow uses COSXL as base image generation before flowing through SDXL lightning for refinement. Follow the notes in the workflow for instructions. Play with the PAG values (I've found success in Input/Output too), test with and without Deep Shrink too, and try out different SDXL prompt styles too.If you make something amazing, please share it on here.NOTE: YOU MUST HAVE THE COSXL CHECKPOINT DOWNLOADED TO YOUR MODELS/CHECKPOINTS FOLDER. FOLLOW THE LINKS ABOVE IF YOU NEED TO FIND IT. USE COMFY-UI MANAGER TO HELP INSTALL OTHER MISSING CUSTOM NODES (RGTHREE, PAG ADVANCED, AUTO-CFG, DEEP SHRINK, ETC.).Workflow produces large resolution files. If you maintain default latent sizing, final images will be 1688 x 2536 in size. File size per image is ~5mb. You have been warned.

CJ's Super Simple High Detail COSXL and PAG Workflow

Added a couple example images with metadata for easy drag-drop comfy-ui reference.

Files with SHA256: eb7f4e4d72b16229a35936467d507a92699e8cc0c7cb6dd60cabd4e8dca79936