search

antgroup / echomimic

EchoMimic: Lifelike Audio-Driven Portrait Animations through Editable Landmark Conditioning
https://antgroup.github.io/ai/echomimic/
Apache License 2.0
3.06k stars 356 forks source link

inference speed is very very poor of echomimic #131

Open Arvrairobo opened 3 months ago

Arvrairobo commented 3 months ago

Do we have any update on inference speed or real time video generation? otherwise this beautiful project is of no use.

i tried g4dn and g4ad both aws GPU family, tried even in 64 GB GPU. it took about 2.5 hours to genreate 45 sec video. about 335 sec/it thats massive.

other day i tried to generate a same 45 sec video on p3 instance which uses v100 GPU and the inference speed was about 125 sec/it

so pratically with this inference speed, this beautiful project is of no use.

Do we have any update on better speed or close to real time generation of videos?

@lymhust i am very thankful to you that you informed on lots of other similar projects that echommic is opensourced but with that speed it is really of no use.

@yuange250 @JoeFannie any one from the author can shed some lights on road map of this project, i can see since last 3 weeks no updates.

Thanks

luyuhua commented 3 months ago

101 same as this issue

WQIANPU commented 3 months ago

我也有这个问题