Innovate Futures @ Benji patreon

Better Consistency With A Modify SVD :)

🕑 Added 2024-07-12 19:24:49 +0000 UTC

Comments

Benjamin Law

Hehe.. I am lookinng forward to Black Forest Ai video model. But AnimateDiff is still a useful tools, it is good tool like a refiner, to fix some morphing parts in ai video.

Benjamin Law

Pony lcm is a good one

Markus Tiedemann

Benji, this is not the best. But we can learn to do the best out of this ;-)

Markus Tiedemann

We have no other option, it runs but its Buggy and its work ;-) In the meantime im play with the new Ponylcm im found https://civitai.com/models/216590/paseer-sdxlpony-lcm-and-turbo-accelerator

Benjamin Law

Technically, I won't downgrade the image model , which is Flux, and lower the resolution for an old motion model. Second, using Animatediff v3 or even hotshotxl it cannot unleash the potential of a transformer architecture image model. If AnimateDiff creator train a new version of motion model for Flux or DiT architecture diffusion models, then that would be another story, and it will be great.

Markus Tiedemann

There is an vid2vid Workflow on Civitai with an Flux Animateddiff Part, very Bugy ;-) https://civitai.com/models/677201/flux-vid2vid-animate-workflow-ft-fluxnf4-flux-controlnet-and-xlabs

Benjamin Law

The VAE , Unet, as we all downloaded the Mimic Motion individual file, that is Unet.

Mauricio Tonon

Curious about the SVD part. what exactly was modified?

Benjamin Law

hehe :)

Crane

wow

Benjamin Law

Oh that is a good idea. I will try this next version of the workflow update. Its like the stay still background that we have in the AnimateDiff v2v.

bruceneely

I'd like to see a way to keep the background in place and have the dancer move in the frame, like a stable camera in real life, wish I could find more control of the camera in general