Software Engineering Courses (SECourses)•10mo ago

is it frame interpolation model?

TimsonOP•3/1/25, 11:05 AM

I bet Kling uses some kind of frame interpolation for perfomance

TimsonOP•3/1/25, 11:07 AM

the problem is - if the model is not trained for reduced fps output it can mess with motion timings

TTimson is it frame interpolation model?

Furkan Gözükara SECourses•3/1/25, 11:07 AM

yes

Furkan Gözükara SECourses•3/1/25, 11:08 AM

https://github.com/hzwer/Practical-RIFE/tree/main?tab=readme-ov-file#model-list

GitHub

GitHub - hzwer/Practical-RIFE: More practical frame interpolation a...

More practical frame interpolation approach. Contribute to hzwer/Practical-RIFE development by creating an account on GitHub.

Furkan Gözükara SECourses•3/1/25, 11:08 AM

i will use this latest one

TTimson the problem is - if the model is not trained for reduced fps output it can mess ...

Furkan Gözükara SECourses•3/1/25, 11:08 AM

well the model already 16 fps

Furkan Gözükara SECourses•3/1/25, 11:08 AM

so we can double and make 32 :d

FFurkan Gözükara SECourses well the model already 16 fps

TimsonOP•3/1/25, 11:12 AM

oh, I didn't realise the measurements are for 80 frames generation (5s at 16 fps). That makes comparsion with Kling even worse as it outputs 150 frames per 4 minute generation

TTimson oh, I didn't realise the measurements are for 80 frames generation (5s at 16 fps...

Furkan Gözükara SECourses•3/1/25, 11:13 AM

ye kling is lead

Furkan Gözükara SECourses•3/1/25, 11:13 AM

but in some cases

Furkan Gözükara SECourses•3/1/25, 11:13 AM

Wan 2.1 made better output i saw comparisons

Furkan Gözükara SECourses•3/1/25, 11:13 AM

by the way when i add double it will be 160 frames for us

TTimson To me bottom line here is - Wan 2.1 14B is not practically usable. It is on par ...

AiInfluence•3/1/25, 11:14 AM

@Dr. Furkan Gözükara this is exactly what. I meant on Patreon chat

FFurkan Gözükara SECourses true but difference may come at fine tuning training

AiInfluence•3/1/25, 11:15 AM

is it even worth the research time when img2 vid exists?

J2_Nedo•3/1/25, 11:16 AM

i saw people train Lora for hunyuan, i hope this comes to Wan too

JJ2_Nedo i saw people train Lora for hunyuan, i hope this comes to Wan too

Furkan Gözükara SECourses•3/1/25, 11:17 AM

there is already Wan training at the library we use

J2_Nedo•3/1/25, 11:17 AM

oh didnt knew that

AiInfluence•3/1/25, 11:22 AM

BTW multi gpu support for wan which devs are ignoring as you said Dr will make it very competitive

AiInfluence•3/1/25, 11:23 AM

imagine 4x A6000 with 10min gen time it would be dope

AAiInfluence imagine 4x A6000 with 10min gen time it would be dope

Furkan Gözükara SECourses•3/1/25, 11:24 AM

yep and cheap too

FFurkan Gözükara SECourses yep and cheap too

AiInfluence•3/1/25, 11:24 AM

how can we push the devs ? maybe you could implement it for us or not possible?

Furkan Gözükara SECourses•3/1/25, 11:24 AM

i am trying to make kaggle work haha

AAiInfluence how can we push the devs ? maybe you could implement it for us or not possible?

Furkan Gözükara SECourses•3/1/25, 11:24 AM

nope i cant

AiInfluence•3/1/25, 11:24 AM

kaggle?

AiInfluence•3/1/25, 11:25 AM

I can't imagine if 50min on H100 how much it will be on kaggle

Furkan Gözükara SECourses•3/1/25, 11:25 AM

trying 1.3b

J2_Nedo•3/1/25, 11:27 AM

Doc i just heard Wan can also create audio

JJ2_Nedo Doc i just heard Wan can also create audio

Furkan Gözükara SECourses•3/1/25, 11:29 AM

they show

Furkan Gözükara SECourses•3/1/25, 11:29 AM

but they didnt publish yet

J2_Nedo•3/1/25, 11:30 AM

ok, so we still need MMAudio

JJ2_Nedo ok, so we still need MMAudio

Furkan Gözükara SECourses•3/1/25, 11:35 AM

currently yes we have it already also

J2_Nedo•3/1/25, 11:36 AM

i know, but you know it also needs an update for the 5000er

Furkan Gözükara SECourses•3/1/25, 11:36 AM

working on fps increase thing

JJ2_Nedo i know, but you know it also needs an update for the 5000er 😉

Furkan Gözükara SECourses•3/1/25, 11:36 AM

i can do now

Furkan Gözükara SECourses•3/1/25, 11:36 AM

good reminding

AiInfluence•3/1/25, 11:36 AM

@Dr. Furkan Gözükara can I do good inference with a 700$ gpu?

AiInfluence•3/1/25, 11:36 AM

like a 4070

AAiInfluence @Dr. Furkan Gözükara can I do good inference with a 700$ gpu?

Furkan Gözükara SECourses•3/1/25, 11:41 AM

it is good as long as fits into vram

Furkan Gözükara SECourses•3/1/25, 11:41 AM

you can also get 2nd used rtx 3090

Furkan Gözükara SECourses•3/1/25, 11:41 AM

should be even cheaper

FFurkan Gözükara SECourses my unified downloader also downloads it

julius•3/1/25, 11:42 AM

ah okay awesome, thanks! Will check!

FFurkan Gözükara SECourses you can also get 2nd used rtx 3090

AiInfluence•3/1/25, 11:43 AM

from Newegg?

AAiInfluence from Newegg?

Furkan Gözükara SECourses•3/1/25, 11:43 AM

depends on where you live

FFurkan Gözükara SECourses depends on where you live

AiInfluence•3/1/25, 11:44 AM

Greece

JJ2_Nedo i know, but you know it also needs an update for the 5000er 😉

Furkan Gözükara SECourses•3/1/25, 11:44 AM

sadly this wont work yet

Furkan Gözükara SECourses•3/1/25, 11:44 AM

no torchaudio for windows for cuda128

Furkan Gözükara SECourses•3/1/25, 11:44 AM

but still testing

Furkan Gözükara SECourses•3/1/25, 11:45 AM

ye no windows. remind me tomorrow

Furkan Gözükara SECourses•3/1/25, 11:53 AM

wow this is really good compare

Furkan Gözükara SECourses•3/1/25, 11:54 AM

is it frame interpolation model?

Similar Threads

Similar Threads

Similar Threads