Stable diffusion thread

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
Discovered new Image-to-Video solution!
Demo showcase: The Wondering Earth 3

Check out the huggingface demo sites.
Colab also available by Camenduru.

 

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
The academic team at Alibaba group is preparing the release of their AnimateAnyone project.
We will then be able to do AI-assisted tiktok dance videos with accurate motions and possibly little interpolation errors. :p
 
Last edited:

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
Cfg and motion bucket adjustments for camera motion.
Eye-blinking tutorial using SVD.
 

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
New paper on animation by NUS & ByteDance!
Colab by Camenduru.

 

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
Just managed to generate one short dance animation using MagicAnimate.
The face and hands are a mess, but the animation looks stable.

So have to use ADetailer to fix the face and hands.
If fixing hands still fail, have to use coloured hand-maps to paint over the terrible hands.

Custom motion sequence is not supported at the moment, not sure if it will be.
The demo has limited motions.

When custom motion feature is available, it will be useful for style transfer and changing clothing.
Then extract the control-net data for grid generation to achieve better overall quality.
 
Last edited:

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
New paper from NUS again!
DeepCache.
Speeds up SD 1.5 by >100%.
Wait for A1111 or ComfyUI to integrate it.
 

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
SVD AnimateDiff Refiner.
It works on 8gb vram GPU.

This method requires 32gb sys ram.
I encountered an error so far during my test, and it is slow.

So a better choice (for me) is to tile-resample and upscale the original output from SVD.
And try to get more stable movements with the camera motion by controlling the Cfg and motion bucket variables.
 

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
New paper on dancing animation by Alibaba AI group!
DreaMoving

Results look very good!
This is in addition to their Animate Anyone paper.
 

doogyhatts

Arch-Supremacy Member
Joined
Feb 13, 2018
Messages
13,334
Reaction score
3,821
I got access into Pika.art.
The output animation is better than SVD but gets slightly blurry.

Output is a 3 second video. I added the reverse manually.
The output can be appended with a 4 seconds extension.

 
Last edited:
Important Forum Advisory Note
This forum is moderated by volunteer moderators who will react only to members' feedback on posts. Moderators are not employees or representatives of HWZ Forums. Forum members and moderators are responsible for their own posts. Please refer to our Community Guidelines and Standards and Terms and Conditions for more information.
Top