thygate / stable-diffusion-webui-depthmap-script

High Resolution Depth Maps for Stable Diffusion WebUI

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

During file naming, seed number not incremented with batch count

graemeniedermayer opened this issue · comments

commented

Actual behaviour
For instance with the prompt building, with a batch count of 2, you'll get output names like:
00675-1600648804-building.png
00676-1600648805-building.png

00677-1600648804-building_depth.png
00678-1600648804-building_depth.png

Expected behaviour
So the 1600648804 in the second depth image name should be incremented by 1. This can be a bit a frustration when creating similar depth maps because the numbering is the same. This also makes it so the depth image won't give the correct information to png_info.

Some of the observations I've made
processed.seed that is used during saving seems to be only the initial seed. There should have some sort of n_iter or batch count index added to it. It would be nice to do something like processed.seed + processed.n_iter ; however, n_iter isn't attached to processed.

Possible solution
Changing processed.seed to processed.all_seeds[count-1] does seem to work. I haven't tested it for every situation.

Yes, that's definitely a bug, thanks for catching this !

p.n_iter is available, but the processed.all_seeds[count-1] seems like the correct solution, -1 to compensate for the initial grid image..

edit: p.n_iter holds the total batch count, so not useful. I will do the same for prompts, change to processed.all_prompts[count-1]

also changed info = create_infotext(p, p.all_prompts, p.all_seeds, p.all_subseeds, "", 0, count-1)

verified correct seeds are now saved in the filename and png_info.

edit2: this works, even when generating one image, because [-1] returns the last element in python.