Place img2tiles.py into /YOUR_SD_FOLDER/scripts
- Go to img2img and load your base image (for now, only tested on 1:1 512x512 images)
- Choose img2tiles from the scripts select
- Adjust tile size (128 makes a 4x4 grid)
- Adjust overlap
- Choose your model and inference settings
- Pictures with highly recognizable patterns gives the best results
- Low steps count with high denoising strength (>0.8) on a sampling like Euler_a, Euler and DDIM outputs great results
- You can use any model you want, even Dreambooth ones
- You can erase tile divisions with inpainting (fun fact: Runway inpainting model doesn't work that great)
Based on sd_upscale.py