Any way while training to generate more than one image?

[enn-nafnlaus]

"Read parameters (prompt, etc...) from txt2img tab when making previews" doesn't really do what it says. You can tell it to make a batch of 8 images, but you'll only get 1. You can tell it to do an X/Y plot, but you don't actually get one. It seems to only read in a couple key parameters.

This matters because looking at a single image doesn't really give you a good idea how training is going - it might be that that particular seed is doing really well or really poorly while most are doing differently. I'd love to be able to generate an X/Y plot or batch of images at each cycle - it'd give a much better sense of how training is going.

Has anyone found a way to do that?

Also, unrelated, but: does anyone else get just a blank preview window during training?

I have to open an image viewer and look int the output directory's images subdir to see the progress. I suspect that the issue is because I use --listen for remote access? Does this happen to anyone else?

[ryanbrackney]

FYI: batch_size definitely should generate more than one image, and if it's not then there's something wrong with your config. (wish I could help!)

[enn-nafnlaus]

It certainly does in txt2img. Not in hypernetwork training.

[ryanbrackney]

ah - an important detail left out of the original post ;)

[enn-nafnlaus]

Did you miss the repeated references to "training" and my quoting of a training parameter (""Read parameters (prompt, etc...) from txt2img tab when making previews")? :)

[CodeExplode]

If you can do a little programming, it looks like you could change how this call is done to just make it create multiple images instead of just one, however the other parts of the code do that. The surrounding code will just read the first image for the previews etc.

https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/modules/textual_inversion/textual_inversion.py#L322

It looks like you could just write

p.batch_size = 4

above processed = processing.process_images(p), to get it to generate 4 previews.

And then change image.save(last_saved_image) to something like

for img in processed:
    image.save(img)

I'm not sure if you'll need to add more seeds or if it will increment automatically.