search

BadToBest / EchoMimic

Lifelike Audio-Driven Portrait Animations through Editable Landmark Conditioning
https://badtobest.github.io/echomimic.html
Apache License 2.0
2.26k stars 263 forks source link

How is 10x acceleration achieved? #60

Closed FacePoluke closed 1 month ago

FacePoluke commented 1 month ago

Thank you for your amazing work. I am curious about how the 10x acceleration is achieved. Can you give me the details?

oisilener1982 commented 1 month ago

i also want to know this. Inference is still the same for me after updating

rammiah commented 1 month ago

i also want to know this. Inference is still the same for me after updating

only pose-driven accelerated model released, audio-driven model is still slow

oisilener1982 commented 1 month ago

Thanks for the update. I mainly use audio driven

LiamTTT commented 1 month ago

It seems to be some kind of improved LCM-Like method.

JoeFannie commented 1 month ago

It is a step-distill technique proposed by our previous work in ECCV2024.

leemengtw commented 1 month ago

It is a step-distill technique proposed by our previous work in ECCV2024.

@JoeFannie great work from the team! Sorry for replying in this closed ticket but just want to confirm is the SlimFlow the previous work the team is using to do the accleration?

wangxihao commented 1 month ago

It is a step-distill technique proposed by our previous work in ECCV2024.

Thank you! Which specific article are you referring to?