for those with 4090 what is the average generation time for I2V 480?
for those with 4090 what is the average generation time for I2V 480?
Mi’ve got 28 minutes
Fi have shown 5090 speeds on video but dont have 4090
Fshown 3090 too 
Falso it depends on
Fresolution
Fnumber of steps
Fnumber of frames
AYep I've got a little faster than this around 20 to 25
AYou can speed up a lot by dropping the resolution but it won't look as nice
MI left thre resolution at 480p and reduced the inference steps to half (25), and dropped the fps to 12 with 60 number of frames and I got 11 minutes
Mand it turned out ok, probably depends on the scene
AI have a 4090 with 16 GB of dedicated VRAM, 64 GByte of RAM on a i9 13900HX and with 9:16 default resolution, 20 steps, 160 frames last night processed 8n over 10 hours. (laptop)
Maybe I am doing something wrong?
Maybe I am doing something wrong?
Fit is long
F10 seconds
AAlso laptop 4090 != Desktop 4090 afaik
AYeah frame dropping is a good option bc you can interpolate them back later using something like flow frames as long as you don't go too nuts
C
Fi tested this on myself
Fresults were horrible :d
Couhh, good to know
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dits not the same one doc, Ive tried it and got good results
AHi.
I would like to ask what is the best Open Source project, that can be installed locally, to create avatars and use them for generating videos, eventually with text to speech capability (voice cloning?) and nice GUI.
Alternatively, what are free or paid version online?
Thank you in advance.
I would like to ask what is the best Open Source project, that can be installed locally, to create avatars and use them for generating videos, eventually with text to speech capability (voice cloning?) and nice GUI.
Alternatively, what are free or paid version online?
Thank you in advance.
Fwhat you can do is
Fgenerate avatar with flux
Ftrain flux
Fgenerate more images of that avatar
Ftrain again
Fthen you can consistently generate it
Fthen you can use wan 2.1 to generate videos
Ffor text to speech i dont have up to date tutorial but you can use elevenlabs
Zshould the wan2.1 take 40 minutes to generate a 5 second clip on a 3090? using the default settings with tiled unchecked
Fstill we are lacking teacache
Fhopefully it will half the duration
Flets try
Flooks even weaker than v express
FFand gave only 4 second output lol
Fgenerating v express
Ayeah only 4 second
Fgithub has long
Fi may make installer
Athank you ,,,, you are the best
Wguys does stable diffusion in google collab still working?
AGitHub
STAR: Spatial-Temporal Augmentation with Text-to-Video Models for Real-World Video Super-Resolution - NJU-PCALab/STAR
Fi plan to make app for this
Ihi ! there is a text to voice model Sesame ai released https://huggingface.co/sesame are you going to create something around this ?
Fyes i plan to but how good it is? i saw today