Keep CFG scale around 4 to 6 for good results. I use DPM++2S Karras as sampling method. Sampling steps usually around 30. For 2.5D look use Rev animated or criarcys fantasy model. For somewhat realistic ones i use deliberate or epicrealism.
Description
FAQ
Comments (19)
how can you do the face amination?
Its done with an extension installed on automatic1111 called Sadtalker. I am thinking about making a tutorial on it.
@Bolly_Diffusion thank you. a tutorial would be great
@Bolly_Diffusion Interested!
@Bolly_Diffusion please do it!
Yes
Found the extension and it is really incredible and scary. Now my characters can talk.... Goes well with a good (local) tts-ai (still looking for one, xvasynth and coqui-ai are not there yet).
@Blockhead Well i used local tts with some tweaks and they work great... still putting it all in one place for tutorial.
how did you make this animation? It is breathtaking
Its done using sadtalker. I am planning on making a short tutorial for that soon.
i will wait for it
@Bolly_Diffusion Is the resolution of SadTalker only 256 x 256 or do i make anything wrong
@nanunana resolution should not be problem if you have 8gb+ vram, it also has face enhancement as default resolution is too low which should clear the problem.
@Bolly_Diffusion Don´t understand this answer. can I set the resolution somewhere? in the manual it says that you can do this in the inference.py but I don't understand how and where
woow.. how did you animate the face? is it free ?
Yes it is free, i am making tutorial for that soon.
That's all I want to know too!
@Bolly_Diffusion Hi, is there any update on the tutorial.. ?
@Dewon You can watch olivio's tutorial for the same, there are too many tutorials on youtube for that now and i did not get time for making tutorial due to personal circumstances. Sorry for keep you waiting. Here is the link to olivio's tutorial: https://www.youtube.com/watch?v=AoIzJWnQw1M
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.