CogVideoX - AI Video Model Run Locally - Too Good To Be True?
HTML-код
- Опубликовано: 13 сен 2024
- In this video where we delve into the AI-generated video models, CogVideoX. Join me as we explore the key features, implications, and examples of this cutting-edge technology, offering insights into its capabilities and limitations.
New 5B model update : • CogVideoX 5B In ComfyU...
In this video, we break down the development of CogVideoX by researchers from Zhipu AI and Tsinghua University, highlighting its advancements in video length, coherence, and motion complexity. We'll also discuss the hardware requirements for running CogVideoX locally and provide a step-by-step guide on installation.
From generating cartoon animations to natural landscapes, we'll review various video examples created by CogVideoX to showcase its performance. Moreover, we'll offer a demonstration of a video-to-video workflow, unveiling the potential for object transformations in videos.
Stay tuned for an in-depth analysis of the AI video model's functionalities and a firsthand look at its outputs. Don't miss out on this comprehensive exploration of CogVideoX and its impact on the AI technology landscape!
Research Paper : arxiv.org/abs/...
GitHub : github.com/THU...
Huggingface Space : huggingface.co...
ComfyUI Node : github.com/kij...
If You Like tutorial like this, You Can Support Our Work In Patreon:
/ aifuturetech
Discord : / discord
Damn, I was hoping the results would be better than SVD. They're almost the same, maybe a little worse. Thanks for taking the time to demonstrate and research this
This is super interesting. This kind of advancement will be important since personally I don’t think Sora will ever be released to the public.
Very high percentage Sora will be not publicly open for use. They worry people abuse it.
@@TheFutureThinker Agreed. They just need to accept the fact any tool will be misused if it can be and let the legal or ethical ramifications be on the heads of the users.
@gnarlymode Unfortunately, people are suing the companies directly
@@AB-wf8ek I get it's an issue, but it seams like they could get around it with their terms and conditions. Kind of like the liability waiver you sign for other things. Use at your own risk. We are not responsible for your misuse. I don't know. I'm not a lawyer
Yes, this is a matter for company. And still AI, how can they define who responsible, etc. its still in discussion.
not commenting often, but it's nice to see this kind of tests even if the result is not good, that's the same work for you. Thank you
In so exited with SOTA realeased by Black forest lab
Yes
Hi, very nice, I tested it, not bad but they need to improve anyway! Thanks!
Switch to Dev-Mode in Manager lets you install the nodes via Manager.
Is there anyway to use a LORA to resolve the issues ?
This has amazing potential!
Wow....
Scroll up to the newer video where I run 5B parameters vereion.
you said you would do a video about face video refining using animate diff. WIll you make it ? I'm really interested in that ! thanks for your content
lol. love the red hat one
How about reviewing VEnhancer in the next video?
Hi what the diffrent between CogVideo2B and CogVideoX-2b?
Thanks a lot for the video and for saving me time. This model is damn terrible and I will not be installing it today, sir.
Wisely move 👍
This could be fun for glitchy stuff
Yes, just for fun. This model cannot tread it seriously.
sadly FLUX dont work on AMD gpu..... on windows, i have 24gb 7900 XTX GPU but cant use it cause it dosnt support....
I'm wondering when we can combine Flux with some video models
You can rigth now connect the outout if your fllux to sdvxl or tooncraft with IPAdapter and control net to generate last fram with sdvxl
How about video crafter.
great
and hey....can i run this on 12 gb vram and 16 gb ram?
Yes , it can this time. But don't expect good quality. You will have surprise.
Can 16G graphics card GPU be used?
Supposed yes
Will it work on 8gb vram invidia 4070
@@FSK2 4070 has 12gb vram
@@kironlau thanks bro
Okay, forget about this one. Next 😐
Wisely move😎👍
Dude! Not knocking the tech but these generated videos got me dying over here!! 🤣. That.........monkey........I'm done!!
😂😂😂🤣🤣
ITS LIKE 2023 ALL OVER AGAIN HAHA
The results are poor in this video because he is only sampling at 25 steps. You really need 50 steps, and more may be even better. Also, either he is using the 2B model (which is poor) or not using the right scheduler setting (the 5B model, which is far better, should be used with DPM).
That is the time when 2B available, check the date. And newer video of 5B which I did , the result better. Its not the matter of steps, scheduler or whatever excuses.
Its just 2B didn't perform.
😂😂 the red hat generate video
hehe :)
I think I've seen more impressive and longer videos with animate diff.
Thats the point👍
bro...lets take a break from FLUX, 1 like by me dude! haha
😂😂😂 have fun testing stuff
Oof... Yet another half-baked local AI video model.
It's rather sad that AI-as-a-service models are so much superior than anything available for use locally. he big problem I have is that the online AI services are exceedingly stingy (even the paid plans)! They generate minutes of waste (sometimes fail in a disturbing manner), are uncontrollable and more miss than hit, which ultimately makes them useless
unless you're willing to shell a hundred dollars every month. 😢
disappointing