[Issue]: Live previews not working

[SAC020]

Issue Description

Since last week (at least) live previews are no longer being generated, at all

Attaching sdnext log (no errors)

sdnext.log

and browser console log

127.0.0.1-1737553840934.log

Version Platform Description

15:40:21-840456 INFO Python: version=3.11.9 platform=Windows bin="C:\ai\automatic\venv\Scripts\python.exe"
venv="C:\ai\automatic\venv"
15:40:22-111048 INFO Version: app=sd.next updated=2025-01-16 hash=e22d0789 branch=master
url=https://github.com/vladmandic/automatic/tree/master ui=main
15:40:22-824446 INFO Platform: arch=AMD64 cpu=Intel64 Family 6 Model 165 Stepping 5, GenuineIntel system=Windows
release=Windows-10-10.0.26100-SP0 python=3.11.9 docker=False

Relevant log output

Backend

Diffusers

UI

Standard

Branch

Master

Model

FLUX.1

Acknowledgements

• I have read the above and searched for existing issues
• I confirm that this is classified correctly and its not an extension issue

[vladmandic]

does the issue exist on dev branch?

[SAC020]

It does, live preview shows a black box.

This time I've generated 2 images in a row in order to also make a screenshot of the UI.

sdnext.log

127.0.0.1-1737559242984.log

17:12:19-455670 INFO     Python: version=3.11.9 platform=Windows bin="C:\ai\automatic\venv\Scripts\python.exe"
                         venv="C:\ai\automatic\venv"
17:12:19-696612 INFO     Version: app=sd.next updated=2025-01-22 hash=9f3a34d0 branch=dev
                         url=https://github.com/vladmandic/automatic/tree/dev ui=main
17:12:19-698958 DEBUG    Branch mismatch: sdnext=dev ui=main
17:12:20-028913 DEBUG    Branch sync failed: sdnext=dev ui=dev
17:12:20-776936 INFO     Repository latest available e22d0789bddd3894364b0d59a4c9b3e456e89079 2025-01-16T16:54:17Z
17:12:20-794042 INFO     Platform: arch=AMD64 cpu=Intel64 Family 6 Model 165 Stepping 5, GenuineIntel system=Windows
                         release=Windows-10-10.0.26100-SP0 python=3.11.9 docker=False

[SAC020]

With TAESD it seemed like it generated a preview at step 38/40

sdnext.log <== 3 generations in a row

127.0.0.1-1737559751386.log

On second generation with TAESD, it started a preview at step 27/40

127.0.0.1-1737559890875.log

On third generation, started at step 30/40

127.0.0.1-1737560042869.log

[vladmandic]

a) taesd is not that lightweight, b) flux is a heavy model, c) live preview is internally run with lower priority so it doesn't slow down main loop.
so unless you have a high-end gpu, this is pretty normal.

[SAC020]

I have a 4090.

"approximate" doesn't seem to work at all, but I will do more retries.

Is it possible to parametrize the priority? I would be willing to make that tradeoff instead of generating blindly.

[vladmandic]

approximate or simple do not exist for flux or sd35, only taesd.

Is it possible to parametrize the priority

not really without massive work and tons of if/then since dealing with torch threads/streams is very gpu specific. if i allow any changes for you, chances are it would completely break on non-cuda environments.

[SAC020]

ok, understood. thank you

[vladmandic]

i'll keep the issue open and i'll see what i can do, i just wanted to set expectations ahead.

[SAC020]

fair enough, thanks!