Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Scheduler bugs and info request #267

Open
ctf05 opened this issue Jan 9, 2025 · 2 comments
Open

Scheduler bugs and info request #267

ctf05 opened this issue Jan 9, 2025 · 2 comments

Comments

@ctf05
Copy link

ctf05 commented Jan 9, 2025

SDE-DPM and DPM give me the exact same results in both generation time and output.

SASolver gives me the error:
sde-dpmsolver++ is not implemented for <class 'ComfyUI-HunyuanVideoWrapper.hyvideo.diffusion.schedulers.scheduling_sasolver.SASolverScheduler'>

Info please:
What scheduler was used before we could choose the scheduler?
Which one is the best (quality and/or time)?

@kijai
Copy link
Owner

kijai commented Jan 9, 2025

Looks like I had forgotten to push some stuff for the schedulers, sde if enabled was not resetting.

The default is the basic Flowmatch scheduler. I find that normal dpm is slightly better than that, and is compatible with everything else. SDE dpm doesn't work with TeaCache, but can be quite strong and give higher fidelity outputs. SASolver is probably overly strong and not even sure if it can work for this, left it there for testing though.

@ctf05
Copy link
Author

ctf05 commented Jan 10, 2025

Looks like I had forgotten to push some stuff for the schedulers, sde if enabled was not resetting.

The default is the basic Flowmatch scheduler. I find that normal dpm is slightly better than that, and is compatible with everything else. SDE dpm doesn't work with TeaCache, but can be quite strong and give higher fidelity outputs. SASolver is probably overly strong and not even sure if it can work for this, left it there for testing though.

Great, thanks. DPM takes twice as long as flowmatch for me. Is this expected?

Edit: nevermind, this was only with heavy blockswapping. It is only a bit longer without that.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants