This is again a very useful, interesting test! I personally would be more interested in the Large version, because it focuses on image quality rather than speed.
Try this settings: Model Sampling SD3 shift on 2.00, steps on 4, cfg on 1,sampler - euler, scheduler-beta. Time for generation with amazing results about 15 seconds.
@@pawelthe1606 thanks and unfortunately now i'm able to use cpu interference , os either turbo sd35 or waiting for a similar lightning 4step lora , then i stuck at cfg 1 , so i ask about what best scheduler and sampler for realistic output
For me the best settings tested for realism (portrait) are: Model Sampling SD3 shift on 2.00, steps on 4, cfg on 1,sampler - euler, scheduler-beta. Time for generation with amazing results about 15 seconds.
Hello, yes. You can use the images from the GitHub repository (link in description). You can drag and drop the image onto ComfyUI to load the workflow.
This is again a very useful, interesting test! I personally would be more interested in the Large version, because it focuses on image quality rather than speed.
Thanks! Yes, I am doing the large version but it's taking longer
Thank you Sharvin for this comparison!
Thank you very much! Thanks for the support!
Amazing work mate
Thanks a lot!
Nice thanks 🙌
Thank you too
would you do the same test but for realistic picture result ? in a new video
Try this settings: Model Sampling SD3 shift on 2.00, steps on 4, cfg on 1,sampler - euler, scheduler-beta. Time for generation with amazing results about 15 seconds.
@@pawelthe1606 thanks and unfortunately now i'm able to use cpu interference , os either turbo sd35 or waiting for a similar lightning 4step lora , then i stuck at cfg 1 , so i ask about what best scheduler and sampler for realistic output
Yes, I planned to experiment with realistic pictures too.
No, please do it in the same video 😜
Hi! Did you try the same with more steps or on non turbo model ?
Hello! Yes, I am doing the large version but it's taking longer. I will post the results once completed.
I use the deis sampler for everything it works for 3.5 too
Thanks for sharing.
For me the best settings tested for realism (portrait) are: Model Sampling SD3 shift on 2.00, steps on 4, cfg on 1,sampler - euler, scheduler-beta. Time for generation with amazing results about 15 seconds.
Thanks for sharing! Will try it.
Отличное видео, спасибочки. Можно вашу схему для Comfy Ui?
Hello, yes. You can use the images from the GitHub repository (link in description). You can drag and drop the image onto ComfyUI to load the workflow.