### OpenAI Sora: 텍스트-비디오 모델 유출과 아티스트의 반발

### 소라 리크: 논란의 여지가 있는 OpenAI의 텍스트-비디오 모델 공개
* 🔍 OpenAI의 텍스트-비디오 변환 모델인 **Sora가 유출**되어 크리에이티브 커뮤니티에서 큰 화제를 불러일으키고 있는 것으로 알려졌습니다.
* 🖼 이 유출은 **Chubby라는 채널 친구**에 의해 처음 알려졌는데, 그는 이 유출을 발견한 **레거시 루머가 허깅 페이스 저장소에 대한 링크**와 함께 유출자의 스크린샷을 공유했다고 언급했습니다.
* 🎥 소라는 지난 2월 텍스트에서 고품질의 동영상을 생성하는 획기적인 모델로 소개되어 **사실적인 결과물**로 많은 관심을 모았지만, 1년이 지난 지금까지도 **대중에게 공개되지 않고** 있습니다.
* ✍ 유출된 문서에 따르면 소라는 일부 아티스트에게 피드백을 받을 수 있도록 제공되었지만, **불과 3시간 만에 갑자기 액세스 권한이 취소**되어 OpenAI의 의도에 대한 우려가 제기되었습니다.
* 💰 유출자들은 성명서를 통해 **OpenAI가 아티스트의 작품을 보상 없이 착취**하고 있으며, 아티스트의 기여를 통해 이익을 얻으면서 **'무급 R&D'로 이용**하고 있다고 비판했습니다.
* ⚖ AI 학습 데이터에 기여한 아티스트에 대한 공정한 보상을 둘러싼 논쟁은 계속되고 있으며, 많은 사람들이 **아티스트에게 자신의 작품이 사용되는 것을 선택하거나 거부할 수 있는 옵션**을 주어야 한다고 주장하고 있습니다.
* 📽 유출된 동영상은 인상적이면서도 불완전한 결과물을 선보였는데, 일부 예시에서는 **모델의 기능**을 보여주는 반면 다른 예시에서는 **결함**을 드러내어 소라가 가능성을 보이지만 아직 개선의 여지가 있음을 보여주었습니다.
* 🗣 크리에이터에 대한 공정한 대우에 대한 요구가 점점 더 커지고 있는 가운데, 이번 정보 유출로 인해 **OpenAI가 앞으로 더욱 투명하고 아티스트 친화적으로 변할지 궁금**해하는 사람들이 많습니다.

소라의 유출 사건은 AI가 제작한 콘텐츠에 대한 아티스트의 기여도 미지급과 기업의 착취에 대한 우려를 드러냈습니다.

Oh boy, Sora, the OpenAI text-to-video model, has leaked! I'm going to tell you everything we know so far and I'm going to show you some of the new video examples that we have. Let's get into it. So, I first saw the leak from our channel's

friend Chubby, and he said OpenAI Sora has been leaked, and Legit Rumors was the first to find this, as far as he can tell. So, he screenshotted the note that the leakers released with the leak and he also linked to this Hugging Face repo

called PR Puppet Sora. Now, before I get into it, let me just go back and explain what Sora is. If you watch this channel, I'm sure you already know, but just a quick overview: back in February, we had this "creating text to video" post and a

new model from OpenAI. It was the first time that we saw anything with text-to-video at this quality, consistent elements, physics. There was a lot of speculation that this might be a world simulator, and there was just so much excitement around it. And then nearly a

year later, we still don't have the model. They've been giving private access to a bunch of artists to collect feedback, and I'll tell you why that point is important in a minute. So, just as a reminder, this is kind of the famous video: a lady walking down the street in

Japan, we have woolly mammoths running in the snow, we have this Spaceman, just everything looked so good, so realistic, and we really hadn't seen anything like this before. Now, obviously, we have some open-source text-to-video models today that are similar, but still not quite the

same quality and here's the important point from the blog post today Sora is becoming available to red teamers to assess critical areas for harms or risk we are granting access to a number of visual artists, designers, and filmmakers to gain feedback on how to advance the

model to be most helpful for creative professionals now that has come to bite them in the butt Let me show you what the leak included So this is PR puppet Sora and right at the top it says after 3 hours OpenAI shut down Sora's early

access temporarily for all artists Now unfortunately none of this stuff works anymore but it looks like it did at least for a few hours Also if you scroll down on the page it says here video generated while the tool was up and unfortunately I think these videos were

taken down as well I have some I found on Twitter that people downloaded before it was taken down so I'll show you some of those in a moment but the interesting part why did this leak happen now If we click right here it says Open letter why

are we doing this now This looks like a manifesto This looks like what Anonymous the hacking group might release but here we are Dear corporate AI overlords very ominous We received access to Sora with the promise to be early testers, red teamers, and creative partners However we

believe instead we are being lured into artwashing to tell the world that Sora is a useful tool for artists Now artwashing is an interesting term I had to think about what that meant for a minute but basically what I believe they're saying is OpenAI is trying to

convince the world that Sora is another tool for artists to create. What they're seeing is Sora actually taking artists' work and is able to then essentially replace artists in the future, and they're not paying the artists. And let me keep going, you'll see.

More artists are not your unpaid R&D. We are not your free bug testers, PR puppets, training data validation tokens. Wow, hundreds of artists provide unpaid labor through bug testing, feedback, and experimental work for the program for a $150 billion valued company, while hundreds contribute for free. A select

few will be chosen through a competition to have their Sora-created film screened, offering minimal compensation which pales in comparison to the substantial PR and marketing value OpenAI receives. I kind of agree with a lot of this. This goes back to the question of should

content creators get paid for their work if the model is trained on that data? And I know a lot of you watching this channel disagree with my take on this. I think artists should get paid and at the very least should have the option to give their work for training or not to.

give their work for training, even if they're not getting paid. Then they just don't give their work. Now the counterargument to that is, well, if a person is browsing the web, they see a piece of content, they're learning from it, and it's essentially the same process

with a large language model, or in this case, a video model. But that's not quite true in my mind. These are non-human entities that are ingesting so much data and that are able to, in a lot of cases, replicate exactly what has been created.

by the original artist and the artists are not getting paid at all this entity is not free and open source specifically Sora is owned by a $150 billion valuation company now Sora is not the first model to go through something like this if you remember back when Midjourney V6 came

out a lot of people were able to recreate shot-for-shot images that matched Marvel movies and other IP and you know they did not get permission to take the IP from Disney so here's an example Mona Lisa here's the Mona Lisa and here's a still from the movie Joker

and here it is created in Midjourney and here are a few other examples of very popular IP that can be replicated using Midjourney now a lot of you think like this person Mike thinks Midjourney takes this is protected by fair use I'm not going to get too deep into that

argument I tend to be more on the side of artists should at least have the option to give their work or not give their work now back to the post with the leak denormalize billion-dollar brands exploiting artists for unpaid R&D and P

furthermore every output needs to be approved by the OpenAI team before sharing this Early Access program appears to be less about creative expression and critique and more about PR and advertisement corporate artwashing detected we are releasing this tool to give everyone an opportunity to experiment with what 300

artists were offered a free and unlimited access to this tool we are not against the use of AI technology as a tool for the Arts if we were we probably wouldn't have been invited to this program what we don't agree with is how this artist program has been rolled out

and how the tool is shaping up ahead of the possible public release. Now we still have no notion of when Sora is going to be released publicly. This might have pushed them to move forward the release date. We are sharing this to the world in the hopes that OpenAI becomes more open,

more artist-friendly and supports the Arts beyond PR stunts. We call on artists to make use of tools beyond the proprietary open-source video generation tools. Allow artists to experiment with the ENT guard free from gatekeeping commercial interests or serving as PR to any corporation. We also

invite artists to train their own models with their own data sets. That's a really cool idea. And then some open-source video tools are VideoX, Mochi, LTX, VideoPyramid, Flow. I think the only one I hadn't heard of is Pyramid Flow, so

maybe I need to give that a try. However, as we are aware, not everyone has the hardware or technical capability to run open-source tools and models. We welcome tool makers to listen to and provide a path to true artist expression with fair compensation to the artist. Enjoy some

Sora Alpha testers and then they list some names here. And yeah, look at that. So Borg Maximalist on Twitter says, 'Settle for nothing less than decentralized AI.' That was 7 minutes ago, and he reposted that message so it looks like they are not being anonymous. They are actually

saying who they are. So here is the famous spaghetti-eating video, and obviously, it's still a little bit off, but this is apparently a SORA leaked video. Here we have a cartoon, semi-cell shaded video. Looks amazing. Look at the consistency of the trees and the

movement behind the character. I mean, it's really really impressive. And here's another version of that famous woman walking down the neon street and it looks phenomenal. Look how her dress is moving. Look at the hands, they really do look flawless. And then when she finally

looks up, a little bit of creepiness with the completely black eyes, but overall looks really good. And here's another one, guy walking in the forest, some incredible looking portal thing that he's walking into and you can see the open AI Sora kind of logo or overlay

right there, watermark that is. And it looks like open AI is trying to quell the leak as much as possible. So Chubby says open AI tries to contain Sora's leak, but they can't, the damage is done. What a day. Sora is really leaked and

here we go in the open AI Discord, you have been muted. Reason: Sora leak topic. So they don't even want anybody talking about it right now. All right, here's another one, a basically Coca-Cola commercial kind of a famous one, polar bear with a can of Coke, looks fantastic.

And another one, the Eiffel Tower in Paris, bunch of tourists looking around. It's raining, really really nice. And let's go through some more. Here's a fire at what looks to be an office building or a factory. The fire looks pretty good, the smoke looks incredible. The factory

just looks a little bit off. Some of the angles, some of the shadows, the reflections, they look a little off, but overall incredible. Here's another one, anime girl walking down the street, really nice, very very cool. Yeah, I love it. It looks so cool. Here's another one.

Teddy bear balloons all around. Feels like this is something we've seen before, but maybe not. And even more, we have a city, cars going by. So if you look at this white car right there, you can kind of see it transforms into a different lane. So some issues. And with these books, we

see some flashing here. There are two pencils. Let me go back for a second so you can see right there, there's kind of two pencils, one main pencil, one weird second pencil right there. But everything looks really good. And here's a picture of a

galaxy spinning. Very cool, very dizzying. And here's one, cat riding a balloon. Weird angle, but it still looks good. Kind of a fisheye lens. Yeah, the cat looks good. And here's one that people are saying is the absolute best in terms of

quality. I mean, the bird's movement, the feathering, the eyes, everything. It is flawless. If I saw this online and I did not know it was AI generated, I would not be able to tell the difference. This is by far the best one. I agree. And so a

couple other notes from this leak, we have some information on the payload sent to the model to get the actual generations. So you can say the number of frames, you can have the number of variants. Obviously, you have the prompt, video gen as the type, height and width

of the video, style. So you can include a style, which is kind of cool, in paint items and model turbo, which means there is a fast generation version of Sora. So here's an example payload. I mean, really everything seems to have been leaked. All right, we have another example. We have

Two puppies playing and this one does not look good. The puppies are transforming mid-jump, very awkward. You can see they go from head to tail, ready and there it is, very weird looking but fun. And here is a girl playing with a can of Coke, so it looks cool, a little

awkward but looks cool. And here we go, another one. Somebody writing a book with a fire in front of them. The light reflections on the face look decent, not great. The rest of the room should be lit up a little bit differently. Here's another one from the leak, a woman kind

of dancing around, moving, posing for the camera. We have some weird lettering over here. Doesn't look like any language I'm familiar with, maybe a little Russian, but I'm not sure. And yeah, looks really good, very consistent movements. The hair looks good too bad we can't see the hands and

here's another one. It's a pirate cat transforming, writing what looks to be like a Roomba or something, jumping around. And here is a monkey playing with what looks to be kind of a puzzle or a Rubik's Cube type thing. So thank you to Cole

for posting a lot of these examples. Really appreciate it, Cole. I'm going to go ahead and follow you right now. So yeah, that's all we know so far. I'll do another update if we find out more information, but Sora looks to be leaked and I wonder if somebody grabbed the

actual raw model. I'm not sure about that yet. I think probably not because that is hosted at Hugging Face and probably behind an API, but maybe we'll find out. Do you think those artists were right to leak Sora? And obviously, it wasn't going

to be leaked long if it's behind a hugging face page now somebody leak the weights which I don't think they did everybody would have it already and you just need to figure out how to run it but yeah let me know what you think in

the comments I think it's cool we got some more information finally I think it actually will push up the release date of Sora give a little bit of pressure to OpenAI I kind of agree with some of the takes that the artist made but I also am

trying to understand the free use argument from the other side if you enjoyed this video please consider giving a like and subscribe and I'll see you in the next one

OpenAI's Secretive Project "Sora" LEAKED!

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

Matthew Berman

Leak of OpenAI's Sora text-to-video model and initial reactions

Announcement of the Sora model leak and preview of upcoming video examples.

The leak originated from a Hugging Face repository called 'PR Puppet Sora'.

Brief overview of Sora's capabilities and initial release in February.

OpenAI granted private access to artists for feedback, which led to the leak.

OpenAI temporarily shut down Sora's early access after 3 hours.

Reasons behind the leak: Artists' open letter and accusations of exploitation

Analysis of the leakers' manifesto, criticizing OpenAI's use of artists' work.

Accusations of 'artwashing' – OpenAI using artists' work for PR without fair compensation.

Artists' concerns regarding unpaid labor, bug testing, and minimal compensation.

Discussion on whether content creators should be paid for data used in model training.

Comparison to similar situations with Midjourney and intellectual property concerns.

Leakers' demands and call for open-source alternatives

Leakers' statement regarding their motivations and lack of opposition to AI in art.

Criticism of OpenAI's rollout of the artist program and the tool's development.

Hope that the leak will lead to a more open and artist-friendly approach from OpenAI.

Call for artists to utilize open-source tools and train their own models.

Mention of several open-source video generation tools.

Examples of leaked Sora videos and their quality

Showcase of various video examples generated using the leaked Sora model.

Discussion on the visual quality and consistency of the generated videos.

OpenAI's attempts to contain the leak and silence discussions on Discord.

Information on the model's parameters and payload structure.

Examples of videos with noticeable flaws and inconsistencies.

Speculation on whether the raw model was leaked and potential future implications.

Question to the audience about the artists' actions and the fairness of the leak.

The presenter's opinion on the situation and potential impact on Sora's release date.

Call to like, subscribe, and watch future updates.