Stable diffusion thread

[doogyhatts]

 

[doogyhatts]

Ok so TensorRT is no longer working now because it requires too much vram.
Cannot export dynamic engine due to failed onnx conversion.

 

[doogyhatts]

Stable Cascade Stage C. Not sure if it will run on 8gb vram.

 

[doogyhatts]

 

[doogyhatts]

 

[RyanJ]

Very naise!

This kind of prompts unfortunately will infringe Firefly guidelines ...

 

[doogyhatts]

Talking & singing facial animation.

Much better than Sadtalker and Vividtalk from Alibaba research group.
https://humanaigc.github.io/emote-portrait-alive/

 

[zheng]

 

[doogyhatts]

SUPIR upscaler.
https://github.com/Fanghua-Yu/SUPIR

support for ComfyUI.
https://github.com/kijai/ComfyUI-SUPIR

 

[doogyhatts]

 

[doogyhatts]

 

[doogyhatts]

The waitlist for SD3 is now open!

 

[doogyhatts]

Ok so I finally upgraded to the Forge version of SD WebUI.
It does seem that the upscaling process is much faster now.

 

[zheng]

Ok so I finally upgraded to the Forge version of SD WebUI.
It does seem that the upscaling process is much faster now.

using for two week already so far so good everything run faster a little.

 

[zheng]

 

[doogyhatts]

 

[doogyhatts]

Tested the new AnimateLCM-SVD-xt-1.1 model.
Can generate 25 frames video in less than 30 seconds on my local machine (excluding model loading time).
Use cfg 1.0 & 4 steps.
But the success rate is very low for 768x768 and 576x1024 sizes.

 

[doogyhatts]

New animation technique from Tencent AI Lab!

 

[doogyhatts]

I managed to get DynamiCrafter working nicely after using Kijai's ComfyUI wrapper.
Success rate is much higher than SVD, so no more generating a huge number of unwanted garbled animations.
However, prompt adherence is not exactly 100%.
Generation performance is fast for 16 frames on my local machine, takes about 4 minutes.
Using 30 steps and cfg 7.5.

 

[kgluong]

Is it open source like stable diffusions?