Open Source AI video model approaches closed ones - Mochi 1

Mochi 1 by Genmo.
The preview version of this model can generate video from text in 480P resolution.
Discord - / discord
X (Twitter) - / mustacheai
Music by Suno.

[@NoName-br8pb]

5 years later and we will create our own anime

[@camilovallejo5024]

You can do it now, no joke, but if you are talking about just writing a prompt and getting a full anime chapter 5 years seems more than reasonable

[@touma-san91]

Jfc... It'll be 2225 before any of us have PC's with that much VRAM to run it locally.

[@happy-gq2kw]

and even with that video memory consumption, still generating 480p, unbelievable. But we have compression technologies, I think it is fp32 or fp16, a variant of say q4 if it will be possible will consume more than 2 times less, although still need to have a server.

[@snintendog]

VRam Size... I ded.
looks like we are seeing the New direction GPUs will take soon for Consumers. As it is VRMA has been sorely lacking for Decades. and Since CPU and GPU Processing has it its limit I say its time to finally use all that bandwidth FINALLY.

[@TarubOtnis]

Hey man my Gemini 1.5 pro didn't generate responses so I thought I could jailbreak it but when I look at the menu there's no jailbreak option please help i helpi ((i just want to make Gemini 1.5 response he always response with dot "..."*

[@cannonball8885]

Please do video on how to use local vision model with screenshare in Silly tavern. Not sure its possible yet, nonlocal is possible though :(

[@NoidoDev]

I still think for most scenes with characters it would be better to use a game engine, where AI helps a lot with the generation and animation.

[@serupithserupith1638]

Почему ИИ так трудно с аниме...

[@TheHasder01]

320 GB of VRAM???
I think you meant 32 gb vram?

[@NoName-br8pb]

No, 4 H100 = 320 GB of VRAM

[@unQni-7]

320? lol