They also created ghidra! Probably second best

To train a diffusion model that only outputs one image with difference is I think not possible you could do an image to image and then fix the seed so you would get a consistent result and then picking the nearest result that is nearly an identical copy

Sometimes my brain does a little funny with me and I’m not sure if I like it or not… I just didn’t realize this at all