Same seed

(See the previous demo page for an explanation of what a seed is.)

The same seed used in combination with identical settings and input should theoretically yield the same image, but there are complicating factors in the actual implementation of the software. For instance, it is standard to use ‘xformers’ optimizations, as they speed up generation times and reduce video card memory usage, which can be a critically limiting factor.

By using the same seed, it is easy to compare the effects of different settings on images. Here is a grid showing the ‘same’ image across four different seeds (columns) with four different samplers used (rows).





By looking down a column you can see how the same seed is interpreted by a different sampler. And by looking across rows you can see that the difference between samplers can be much less than the difference between seeds. Adjusting the sampler seems to lead to different images within a relatively small set of latent imagery (so, ending up in the equivalent end destination, but within different universes, each with their own take on reality).