Model DID NOT get leaked. artists DID NOT have access to model. They had API ACCESS to try SORA unlimited and free. They "leaked" this API KEY only, NOT the whole SORA model
It is not PR. OpenAI doesn't want the public to access it yet, they will still be working on red team, and they are not being paid _for literally making adverts for them._
Kling 1.5 and HailuoAI are already quite ahead, then there is Mochi and Cog on the open source side, Sora is completely unnecessary at this point. Also if the model leaked and someone downloaded that it then it most likely is still somewhere on torrent sites, you really can't remove stuff from the internet once it's unleashed.
I like Olivio's videos and find his channel both entertaining and informative, but I think here he could have done a bit better vetting his information. The model was never leaked - API access was. A pretty big difference really!
@@rotggaming4613 I can run MoChi on a Nvidia 1080 Ti 11gb, but it takes 60 minutes at default settings. LTX I think takes me 2 minutes on the same settings. So 30x faster. 1 hour for a clip for mochi is impractical, because if the clip is worthless, you wait another hour. 5 bad clips is 5 hours wasted time. So while you don't technically NEED a 4090, Mochi at default settings is impractical. LTX is fast enough to be practical on my GPU, 5 attempts takes about 10 minutes. Bad outputs at 2 minutes is not too bad.
no longer misleading, people would agree with that now "modelName leaked" (that is only api available) You are not assuming model until it says "modelName model leaked"
You don't have a separate light or enough distance from your backgroumd, also you will want to do a two pass for that chroma key, the first removes the main background, the second is to finetune the edge detail and colorspill
@@FusionDeveloper VRAM is memory, that's what the M stands for - Video Random Access Memory. I don't think it will ever be available for local use either, at best it'll be behind some kind of paywall.
No, an API key(s) have been leaked, intentionally. Obviously artists do not have access to model files, and leaking an API has much less legal implications, since it cost a lot of money toake that model.
Welcome to 202X where official launches are just beta tests, no matter what it is, games, CPUs, everything is same, we are just been used for error testing.
While I do understand the point the artists are making here I think they are undervaluing what they receive by being the first to get access to this new AI. They are going to be the very select few artists validated by OpenAI to be the experts in this technology globally. They are learning how to leverage this cutting edge technology before anyone else this is can quickly boost their position in the global art market.
The problem I have with this is that people are just going to abuse it instead of using it to improve their own skill. Like for example lazy game developers. It has already happened with voices and pictures, and now its going to happen with animations as well
Skill will always triumph, no matter what, and using these technologies as another tool in the shed, is the way to go. Otherwise your gonna be left in the dust
That has always been the case, whether it's photoshop, deepfakes, AI .. there will always be people who abuse stuff, but they will always be a small minority of ppl.
everyone is responsible for their own actions, also thats such a lame argument, theres car accidents daily, ban the cars, people are abusing pills, ban the pills. meh...
@@mvstermlnd 100% agree. It's no different than banning people from using hammers, or insisting all knives sold must be dull - because some people hurt others with them. XD
Llama, not Ollama¹. SDXL 0.9 too, and NovelAI before that. In this case though, the model wasn't leaked, just access to it. ¹ Llama is the model, Ollama is a tool to run LLMs in a Docker-like fashion.
Model DID NOT get leaked.
artists DID NOT have access to model.
They had API ACCESS to try SORA unlimited and free.
They "leaked" this API KEY only, NOT the whole SORA model
Good point, i changed the title
Leaks in general are great for people. This is not a model liberation, just a PR move.
It is not PR. OpenAI doesn't want the public to access it yet, they will still be working on red team, and they are not being paid _for literally making adverts for them._
Kling 1.5 and HailuoAI are already quite ahead, then there is Mochi and Cog on the open source side, Sora is completely unnecessary at this point.
Also if the model leaked and someone downloaded that it then it most likely is still somewhere on torrent sites, you really can't remove stuff from the internet once it's unleashed.
I like Olivio's videos and find his channel both entertaining and informative, but I think here he could have done a bit better vetting his information.
The model was never leaked - API access was.
A pretty big difference really!
LTX video is going under almost everyone's everyone's radar. It's super fast.
On a 4090, it takes longer to watch the output than to render the video.
MOchi requires great card , Tesla Nvidia or 4090
@@rotggaming4613 I can run MoChi on a Nvidia 1080 Ti 11gb, but it takes 60 minutes at default settings.
LTX I think takes me 2 minutes on the same settings.
So 30x faster.
1 hour for a clip for mochi is impractical, because if the clip is worthless, you wait another hour. 5 bad clips is 5 hours wasted time.
So while you don't technically NEED a 4090, Mochi at default settings is impractical.
LTX is fast enough to be practical on my GPU, 5 attempts takes about 10 minutes.
Bad outputs at 2 minutes is not too bad.
@rotggaming4613 Sure, we still have Cog which can work on 12-16 VRAM.
The title is misleading. someone just leaded the API so a few people tried it out!
no longer misleading, people would agree with that now "modelName leaked" (that is only api available) You are not assuming model until it says "modelName model leaked"
You don't have a separate light or enough distance from your backgroumd, also you will want to do a two pass for that chroma key, the first removes the main background, the second is to finetune the edge detail and colorspill
I will only be impressed from a "gaming scene" if the character can go the same way twice...
yes, that might be a huge problem for now
the problem here is how many memory it will use .........
VRAM, not memory, but it likely will never be released to run on local GPUs.
@@FusionDeveloper VRAM is memory, that's what the M stands for - Video Random Access Memory.
I don't think it will ever be available for local use either, at best it'll be behind some kind of paywall.
I kind of think all of the really good models we see hosted by companies use way more vram that a local use can have. hopefully that changes over time
No, an API key(s) have been leaked, intentionally. Obviously artists do not have access to model files, and leaking an API has much less legal implications, since it cost a lot of money toake that model.
The public will specifically want the turbo model because it will be slightly a lot less expensive.
Sora had to be better with dynamic camera? Downgrade?
I think that baby farted underwater
an DALL E model leak would be more useful than this 😂💔
Welcome to 202X where official launches are just beta tests, no matter what it is, games, CPUs, everything is same, we are just been used for error testing.
It's green banana time ;)
Don't understand the excitement now? Sora was already known, they just made a turbo version.
the excitement is the leak and this drastic action by the artists
hahhaah nice shirt
This model was not that impressive.
I think for a turbo model it was very nice. But of course it can always be better ;)
Yeah.. I really don't wanna invest time in learning everything about video AI and how to use it as long as models are as janky as they are right now.
Caf tail noclips Lol
While I do understand the point the artists are making here I think they are undervaluing what they receive by being the first to get access to this new AI. They are going to be the very select few artists validated by OpenAI to be the experts in this technology globally. They are learning how to leverage this cutting edge technology before anyone else this is can quickly boost their position in the global art market.
The problem I have with this is that people are just going to abuse it instead of using it to improve their own skill. Like for example lazy game developers. It has already happened with voices and pictures, and now its going to happen with animations as well
Skill will always triumph, no matter what, and using these technologies as another tool in the shed, is the way to go. Otherwise your gonna be left in the dust
That has always been the case, whether it's photoshop, deepfakes, AI .. there will always be people who abuse stuff, but they will always be a small minority of ppl.
everyone is responsible for their own actions, also thats such a lame argument, theres car accidents daily, ban the cars, people are abusing pills, ban the pills. meh...
@@mvstermlnd 100% agree. It's no different than banning people from using hammers, or insisting all knives sold must be dull - because some people hurt others with them. XD
@@mvstermlnd Which argument are you referring to? Nobody here mentioned 'banning' anything except for you and a reply to you.
Nice
Models leak surprisingly often.
I think the original Ollama model was also leaked
Llama, not Ollama¹. SDXL 0.9 too, and NovelAI before that.
In this case though, the model wasn't leaked, just access to it.
¹ Llama is the model, Ollama is a tool to run LLMs in a Docker-like fashion.
what the duck you talking about?
Duck knows.
Are u for real ?? maybe it's fake , xD
Of course hes real.