How to improve video quality / sharpness in your workflow (v9 NSFW)
This is fantastic. For real. No idea how you did it !
I am using the suggested workflow with a WAN2.1 lora at 4 steps, CFG of 1, and Euler - the videos are great, but I'd really like to improve the image (video) quality. There is a lot of noise and blurring.
Increasing steps and CFG does not seem to help, not does adjusting the lora.
Is this jus the quality trade-off that we have to live with using the NSFW model, or could someone share some secrets with me ?
Thank you !
I'd recommend starting with the recommended sampler, which isn't euler. Try euler_a with the beta scheduler. Also, try using the "base" all in one to see if the noise goes away, or remove all extra LORAs completely to see if it is a LORA is at fault.
Thanks a lot for your response
@Phr00t
- I'm still experimenting !
I have four of the AIO versions in my folder, the first one, V5, V9, and NSFW V9.2 - Using the same workflow and with everything else constant, I get the best image quality (clearest video) out of the first version.
The NSFW version is definitely better for action, but I have found that it struggles under low lighting conditions (dimly lit).
And when the model does not know what to do, there are some pretty funny variations LOL.
Thanks again for your efforts !