This i2i workflow was originally uploaded by Levpat but they seem to have disappeared. The workflow uses an SDXL realism checkpoint with a tile controlnet to upscale the image in pieces. Don't be intimidated by the huge spaghetti nightmare. Just set the input image on the left and ignore the rest. I recommend using IntoRealism SDXL by enzino and no prompts. Also a semireal model for the input image works best. It's not a super intelligent workflow (no regional prompting) but its beautiful and fast.