Running a batch of select seeds, skipping the bad ones

[kwikmn]

Sometimes, with certain models, I can only make a prompt about 20% successful, i.e. only 1/5 images come out decent.

What I would like to have is the ability to run a huge batch text2img, then take the seeds of only the successful images in this batch of low-res renders and then feed them back as a list of seeds on which to do highres.fix upscaling at double or triple the resolution.

Is there a way to make Auto1111 do this? I would manually have to delete all the bad images, of course, but can you feed a list of seeds that has skips in it?

Something similar, of course, is to feed the machine a batch of images on which to do img2img, but that's not the same thing. I want the seeds to be the same as the images in the low-res batch.

[cgessai]

If you don't get a more authoritative answer, I'm just going to leave the following because it'll likely (I hope) be better than nothing: For somewhat similar reasons I had introduced a feature request in the powerful randomizer extension 'sd-dynamic-prompts' which is installable from within SD, itself. The GitHub for that project is here. In that discussion, the maintainer (adieyal) mentioned there were similar requests with links and a guy named stysmmaker seemed to indicate that he had produced a script which might allow for something. Whatever that something is, is pretty oblique to me but here's the project stysmmaker made and references in that thread. Maybe more information is in the script that on the GitHub page, IDK.