Inportation of image metadata from process not reflecting accurate sampler.

[djwashout]

When I drop a DPM++SDE image in to the process tab and then send data to text it will display DPM++SDE in sampler, however images produced are UniPC.

[vladmandic]

pls upload the image used to reproduce this.

[djwashout]

NSFW okay?

[djwashout]

This was produced on Google Colab, Version: app=sd.next updated=2023-11-03 hash=dc79cec1
Trying to change the sampler to DPM ++3M EDM produces static. Other samplers produce a similar but not exact image.

[djwashout]

So this would happen occasionally when i would drop images over form A1111 in to Vlad, id have to make sure to select the right sampler in the drop down, however i don't see the option for DPM++SDE so i can not take corrective action..

[vladmandic]

i've noticed an issue with negative prompt parsing, that is now fixed.

the image was generated using sdnext with backend=original and now you're using backend=diffusers which has different schedulers.
scheduler dpm++ sde does not exist as such as neither dpm sde can have **++ solver **nor dpm++ can have sde solver.
closest you can get is to set dpm++ 1s with sde solver.

note that solver is set in settings -> schedulers and it modifies behavior of currently selected scheduler.

[djwashout]

Wow thanks! I noticed sometimes when I would load an image into process and send it to text it will not load the negative prompt and i would just cut and paste it in myself.

Also i have changed back to backend=original and i now have DPM++SDE for sd1.5 models. I am now able to reproduce the image.
Sorry for my lack of understanding regarding samplers, schedulers, and solvers. i'm trying to read up on it all.

Thanks Vlad. You do have a far superior product here. It has more creativity and produces amazing images. I have always preferred it over a1111.