31-second audio -> a total processing time of 2617.4 second / 43 min VS realtime

[johndpope]

the key difference between VASA - and V-express from architecture POV - is the use of Megaportraits / resnets to do warping without needing keypoints.

I am attempting to recreate Megaportrait codebase here - https://github.com/johndpope/MegaPortrait-hack
There's some audit on flops using profiler -
johndpope/MegaPortrait-hack#39

is it just a case of ripping out the keypoints in this repo to make inference faster?

[dCodeMaestro]

Interesting. Does V-express yield similar results quality as vasa-1 ?

[FurkanGozukara]

Interesting. Does V-express yield similar results quality as vasa-1 ?

i have full tutorials check them out

76.) Free - Local - PC

https://youtu.be/xLqDTVWUSe

V-Express: 1-Click AI Avatar Talking Heads Video Animation Generator - D-ID Alike - Free Open Source

77.) Free & Paid - Cloud - RunPod - Massed Compute - Kaggle

https://youtu.be/GXBiqJOc9FE

V-Express 1-Click AI Talking Avatar Generator - Like D-ID - Massed Compute, RunPod & Kaggle Guide