Am I the only one who caught an AI image lifting a watermark straight from one of my old scans?
I was running prompts for a woodland scene last Tuesday and the model spat out a fern that had my faint signature loop baked right into the fronds. I cropped that corner, ran it back with a stronger negative prompt and a bigger canvas, and the third batch came out clean with no stray lines. Has anyone else spotted their own old uploads leaking into generated pics, and what did you change in your prompt to cut it out?