OpenAI Sora: 새로운 AI 동영상 생성기 개요 및 기능

### OpenAI, 소라 출시: AI 비디오 생성의 새로운 시대
* 🎉 OpenAI가 드디어 기대를 모았던 **AI 동영상 생성기 모델인 Sora**를 출시하여 그 기능을 살펴보고자 하는 커뮤니티의 열기를 불러일으켰습니다.
* 💰 ChatGPT 유료 계정이 있는 사용자는 **월 20달러의 플러스 요금제로 50개의 동영상을**, **월 200달러의 프로 요금제로 최대 500개의 동영상과 고해상도 및 동시 생성 등의 추가 기능**을 제공하는 소라를 이용할 수 있습니다.
* 🌍 그러나 현재 OpenAI가 해결하고 있는 규제 문제로 인해 **영국이나 EU에서는 아직 Sora를 사용할 수 없습니다.**
* 💻 사용자 인터페이스는 **직관적인 대시보드와 동영상 정리, 프롬프트 필터링, 설정 관리** 등의 옵션이 있어 크리에이터에게 친숙합니다.
* 🎥 초기 테스트에서는 서버 트래픽이 많아 일부 동영상 생성에 **11분 이상**이 걸리는 등 엇갈린 결과가 나왔지만, **롤러스케이트를 탄 원숭이**와 같은 특정 결과물의 품질은 이전 AI 동영상 생성기에 비해 크게 개선된 것으로 나타났습니다.
* ⚠ 많은 사용자가 동영상 생성 중 지연 및 예기치 않은 오류를 경험하고 있으므로, **서버 부하가 줄어들 때까지 기다렸다가 Sora를 사용하는 것**이 좋습니다.
* 📈 몇 가지 어려움에도 불구하고 **2023년 3월 이후 AI 동영상 기술의 발전**을 보여주는 인상적인 사례가 특집 탭에 다수 공유되는 등 Sora의 잠재력은 분명합니다.
* 🚀 사용자들이 Sora의 기능을 탐색하고 프롬프트 기법을 개선해 나가면서 **AI 동영상 생성의 미래는 밝아 보이며, 내년에는 더욱 큰 발전**을 기대할 수 있을 것입니다.

OpenAI의 Sora는 현재 높은 수요와 긴 대기 시간에도 불구하고 뛰어난 기능을 제공하는 AI 동영상 생성에 혁신을 불러일으켰습니다.

Well, the day is finally here! We've been hearing about this forever, but OpenAI finally released Sora, their AI video generator model. I'm super excited to get to play with it. I haven't personally been able to use it yet, although it did

take so long to come out that all of the other AI video generators have seemed to improve and get better, and almost got to the level of Sora. But from what I saw in today's demos, Sora still looks like it's got a little bit of an edge on a lot of

the other tools. So let's dive in and put it through its paces. The thing most people are probably going to be asking themselves is how much does Sora cost to use? Well, if you've got a ChatGPT paid account, you can use it with your account.

So, if you're on the $20 a month Plus account, you get up to 50 videos per month, which is 1,000 credits. You can generate at 720p. If you're on their new $200 a month Pro plan, you can generate

up to 500 videos a month, or 10,000 credits. You also get unlimited relaxed videos, so I'm assuming that's a slower-to-process video generation. You can generate up to 1080p resolution and 20 seconds. You can have five concurrent

generations, so you can have a whole bunch of generations going at once. You can actually download the videos without the watermark. So if you're on that $200 a month plan, you get a bunch of extra features, but if you're on the $20 a

With the monthly plan, you do get to use it as well. I did upgrade to the $200 a month plan because, well, I make YouTube videos about AI, so I figured I should have access to as much as I can. The other thing to note about Sora is that it is not available

in every single country yet. They do have a list of countries that it's available in. There are quite a few countries where it is available right now; however, it is important to note that it is not available in the UK or the European Union yet. They're trying to sort out

you know, all the legalities of getting it live over there because they have a bit more strict AI-related policies in those areas. So let's go ahead and try it out. When I log into the dashboard here, I have a fairly familiar, intuitive-looking dashboard. Over on the

left menu, we've got our recent videos, we've got featured, we've got saved, we've got our library: all videos, our favorites, our uploads, and it looks like we'll be able to organize by folders, which is pretty handy—something I wish ChatGPT actually did. Up in the top right here,

we've got filters to filter by prompts, storyboards, remixes, blends, and loops. We've got some filter options by aspect ratio and square. I don't totally know what that means yet. We've got our activity, and then we have our little profile and settings dropdown here.

But clicking into settings, we've got options for username, email, themes. You can decide whether or not you want your videos published to the explore page, and you can choose whether or not you want your content to be used in training. I can click on my plan here; I can see how

Many credits I've got; how many relaxed videos I've got; max concurrent generations; max video duration; max resolution; and the ability to download without watermarks. So I can see all of the features I've got here. Then, down at the bottom, we have our prompt entry

box. I can upload an image. When I go to upload an image, it asks you to agree to a bunch of policies, so let's go ahead and do that. But I'm not actually going to upload an image right now; we'll come back to that in a minute. We've got some

presets here: balloon world, stop motion, archival film noir, cardboard, and paper craft. And I believe in their demo video this morning they actually said you can create your own presets as well. We've got our resolution; it looks like at the moment we have three options: 16x9, 1x1

square, or 9x16 (sort of vertical videos). We've got our resolution. We have the option for 80p, 720p, and 1080p. We can see how much slower each progressively better resolution is. We've got our duration: 5 seconds, 10 seconds, 15 seconds,

and 20 seconds. I believe you need to be on the higher-end plan to go all the way up to 20 seconds. And you can choose how many video variations you want it to generate when you generate new videos: 1, 2, or 4. And then over here we've got the

storyboard option, which we'll have to play with in a minute. But let's go ahead and test some of the videos. I always like to test, when I first use video generators, I like to do a monkey on roller skates because so far none of

The video generators have done a great job with that one. Let's go ahead and leave it at 5 seconds, 480p, just for the speed of generation, just because it's a test. And we'll create the video now. I can see up here it added to the queue, so I've

got one video running, but I get five concurrent videos because I am on the pro account. So let's go ahead and do another one. Let's do a wolf howling at the moon. We'll generate that one now. I've got two concurrent generations.

Let's do a cat eating tacos. You've watched my previous AI video generation videos; you've seen me do all of these prompts before. Now, for whatever reason, one of my videos failed, so a cat eating tacos... there was an unexpected error running this prompt. Let's try that again.

There shouldn't be any problem with that one, a cat eating tacos. Let's go ahead and generate it. This time it appears to be working. Maybe we'll find out... My other videos still appear to be queued when I look up here. I'm going to go ahead and

rename this new folder that I created. I'm just going to call it 'test,' so when I'm back in Sora later, I know that these were just the videos I was using to test with. And my guess is, since Sora was announced today, there are a lot of

people trying to generate videos right in this moment. It appears to be so overloaded, in fact, that a lot of people right now are getting this message: 'Sign-ups are temporarily unavailable. We're currently experiencing heavy traffic and have temporarily disabled sign-ups. We're working to get them back.'

It'll be up shortly, so check back soon. If you do jump over to Sora expecting to play with it and you haven't signed up for Sora yet with your Pro or Plus account, you may see this message. Hopefully, it's gone by the time you try to sign up for Sora.

But some people are running into this issue. One little coder here shared on X that he actually has signed up but was unable to actually generate a video because it says, 'Service at capacity. Please try again later.' I personally haven't run into this issue yet, but I

have gotten a few unexpected errors. All right, so I got my first generation of a monkey on roller skates. Again, it's only at 480p, and you know the quality is really showing. One thing I find interesting about watching the playback of these videos is if I scroll my mouse

across the video, it actually plays it, so I can actually scroll it forwards and backwards. You can see my little finger there on the left video, and just by moving—I'm not clicking or anything—just by moving my mouse across the video, it's playing forwards and backwards. Now, other

than the fact that these videos are 480p, there is no doubt that this is a monkey on roller skates. This is 100% the best generation of a monkey on roller skates that I've seen so far, and of all the video tools that I've tested, it is a

monkey wearing roller skates. They're clearly roller skates. The leg kind of disappears at one moment on the left; right there, it looks like the legs sort of switch places with each other, but it's pretty dang good. But again, this is

480p, so it's a little pixelated looking. I think it's pretty bogged down. It took about 6 minutes just to wait through the generation queue, and again, I'm on the $200 a month plan. I generated at the lowest resolution, the shortest time possible.

and it took 6 minutes just to get out of the queue, and then once it started generating, it took another 1 to 2 minutes to generate. I wasn't paying close enough attention, but it took a good minute or two to generate. So, all in,

I started the generation at 10:29; it was done by about 10:37, so it took about 8 minutes from the time I entered the prompt to the time I got the video. My wolf howling at the moon, so far, is taking even longer than that; it's at

94% right now, but I started the generation at 10:30; it is now 10:38, so so far we're about 8 minutes in on that one. And my cat eating tacos, I'm at about 7 minutes in on this one. So I finally got

the wolf howling at the moon. It took about 11 minutes to generate these, but let's go ahead and take a peek. You can look at both of them at the same time as I scroll across. The one on the right is barely animated at all; you could see a

little bit of the trees rustling, and the wolf moves a teeny tiny bit. The one on the left here, it just looks like a static image that somebody put a parallax effect behind it; it doesn't even look like a video. The wolf's not moving.

All the trees aren't really moving; it just sort of looks like there's a starry background and then a foreground with a rock, a wolf, and trees in the moon. It's sort of like a parallax between the two images. And then again, on the one

on the right, there's like very subtle movement. So, fairly disappointing for waiting 11 minutes to get this, which is only 40p in 5 seconds. I did get my cat eating tacos. This one took about 11 minutes also. So the one on the top is a

cat maybe eating a piece of lettuce or something. And then the one on the right here, the cat's chewing; you can see a taco there. Not too bad. I think it's fairly impressive; I don't have much complaint about this one. It would have

been nice if, like, the face was in the taco eating it, but it's a pretty solid generation. The wolf howling at the moon was disappointing. The monkey on roller skates, I'm actually pretty impressed. I've had a hard time getting a monkey on

roller skates. Let's try an image to video. I'll go ahead and upload an image here. Here's an image that I used for a recent thumbnail of me kind of talking into a mic. Let's just call it 'A man speaks into a microphone'. I'm a little

worried about trying 1080p right now, because if it's eight times slower and the other ones took 10 minutes, that means it's going to take 80 minutes to generate. So I'll go ahead and leave it on 480p.

Just for the demo purposes of this video, ideally, once Sora has been out a little longer, those wait times will come down. But I do think it's getting bombarded today. Let's go ahead and create the video based on that image.

Then, while we're waiting for the image-to-video, let's try some of these presets. Let's do a Balloon World, and I want to test all of these presets with the same prompt so I can see how the preset adjusts the same prompt. Let's do

a pirate juggling watermelons. I'm going to copy this prompt so I can use it over and over again. We're going to use the Balloon World preset; we'll run that one. I'm going to try the Stop Motion preset, we'll try the Archival preset, and

the Film Noir preset. All right, so the video I tried to generate, where it was starting with an image of my face, that one, after about 20 minutes, it said there was an unexpected error running this prompt. Retry prompt. I'll go ahead and click "Retry prompt" here, and it just

brings up a page that is a larger video showing "unexpected error." So, not sure why clicking "Retry prompt" doesn't actually do anything. I don't know if that's a bug or what's going on, but nothing happens when I click "Retry prompt" other than it just opens a page

with the errored-out video on it. All right, so I got the image-to-video to work after that first unexpected error, and here's what it did. So the first one, it has some subtle motion in the eyes, then it kind of zooms in to show

More of the jacket, and then it starts to talk. The second one starts with just a static image, doesn't move, and then moves to a completely different image. So, kind of funky how that worked. I don't really know how to prompt Sora yet; this is

literally my first day using it as well, so there's probably a lot of things I'm doing wrong with prompting. I'm sure there's a lot of people that are going to give some advice on how to actually get better results from the prompts you

give it. But from this image, it basically said, "Cool, I like your image, but I'm going to make you a completely different video." Now let's look at some of the presets we were playing with. Here's a pirate juggling watermelons using that

balloon world prompt. I guess it kind of looks balloony; it more looks like an almost Pixar-style animation, but I could see it being a little bit balloony. Uh, the one on the right, he more like makes one watermelon turn into two

as opposed to juggling, and it looks like he's just sort of making watermelons float. I'm not really seeing the juggling action. It's a decent-looking video, I guess. This one is a stop-motion video, and yeah, that looks like stop motion; it

doesn't look like a pirate juggling in either one. On the one on the left, it just looks like watermelons sort of floating around in the air. On the one on the right, it looks like a watermelon pops out of the water and floats there.

but it does look like stop-motion animation. Then we have archival footage here. This actually is starting to look a little bit more like juggling, but the balls are sort of floating away. You've got the little, like, marks on the side that almost look like film and it

does look like an older archival video, so it has the preset. I can see that the preset is working; the video—the physics of it—is just still kind of wonky. Definitely looks very AI-generated. And now we have our film noir preset of a pirate juggling watermelons. If I hover

over this one, I mean, it's black and white, and I guess film noir because you could kind of see all the shadows and stuff. But once again, it can't figure out the juggling aspect of it. Uh, they don't really look like watermelons either; they

just kind of look like balls, or, you know, bowling balls or something that he's holding. And then here's the final preset, the cardboard and paper cutout version. Once again, not much juggling. The one on the left looks a little more like juggling; the one on the right looks like

he's just kind of turning one watermelon into two and then dropping it on the ground and turning it into one again. But I definitely get the cardboard and paper cutout aesthetic; it just can't figure out how to do juggling yet. This one also

took about 20 minutes to generate, so it is still slow right now. But we can see what the various presets do. Let's go ahead and try the storyboard feature. I'm going to click storyboard, and this is one of the areas where I feel Sora gets

Really powerful is you can actually give it multiple prompts along a timeline and have it sort of work with those prompts. So I want to try to get like a man standing on a cliff morphing into a wolf. So I can come here to like the 2-second marker, click plus, and now I can add another prompt.

For the first one, we want a man standing on a cliff. Let's go with a full moon in the background, and then 2 seconds in, the man morphs into a wolf. Let's see if we can get a man standing on a cliff with a moon in the background, sort of morphing into a wolf, like a werewolf. And we'll go ahead and click create, and that one's now added to the queue.

The other cool thing about this storyboard is you can generate a 5-second video, and if you don't like the last second of the video, you can crop it down to 4 seconds and then have the storyboard generator regenerate that last second. Or you can slide the whole video over and have it regenerate the first second. So it takes the context of the videos that you have and knows how to extend them.

Let's take a peek at the storyboard generation. Now, remember it starts with a man on a cliff, and then the man is supposed to morph into a wolf. Let's check out the two generations. We can definitely see the man on the cliff there. Let's go ahead and hover over it and let's see what the animation does. He sort of crouches down and doesn't turn into a wolf.

The right, it just sort of zooms in on the man and never actually turns into a wolf.  Maybe it just needs a longer generation. Let's see what happens if I just give it a prompt without using the storyboard feature, like 'a man on a cliff'

morphs into a wolf under a full moon.' Maybe if we try that, as opposed to the storyboard feature, I'll get something a little bit closer to what I was hoping this one would do. But this one's just sort of a dude standing on a cliff.

This one looks like it could just be a static image that's zooming in on the image. Here's my full prompt: 'A man on a cliff morphs into a wolf under a full moon.' We can see the full moon, we can see the man, but we're not seeing much

morphing. So, uh, it seems to struggle with a man morphing into a wolf, but again, maybe a more detailed prompt would have worked. And also, I got one failed variant, so it was supposed to generate two videos; it only generated one. Again, I think this is a symptom of the servers

just being completely overloaded right now. And now, instead of generating more videos to show what it's capable of, I'm going to show some of the videos that are under the 'Featured' tab under 'Explore' here, because every time I'm generating a video, it's taking between 15 and 20

minutes right now, which is making the video I'm recording right now take forever. My advice is, if you want to use Sora, I'd probably wait like a week or so until maybe some of the craziness dies down, because right now it is super

It's overloaded; it's taking a long time to generate videos, even if you're on the high-end $200-a-month plan. Also, we'll probably have more feedback and tips on how to better prompt by then, as well, because right now I'm still kind of confused on

how to get something really good out of Sora. You know, simple, basic prompts don't work well, so chances are you have to give it a little bit more detail to really get what you're looking for. But looking at some of the examples in the featured tab, people are getting really

good videos out of this. We've got like a bird on a fence here that looks pretty good, looks pretty realistic. You've got some clouds; we've got like a—I don't know—looks like a clay rocket here. We've got a bunch of colorful dancing poodles

like River, and this like fantasy-looking world; somebody looking out over a city skyscraper; a monkey in the smoke; driving through some trees here; a ladybug; a man walking through what looks like Tokyo or something; we've got a ship on the ocean; an elephant walking in the snow

here; a cat on a skateboard; a Viking walking in the snow. I mean, the stuff that's showing up in the featured videos is really, really impressive. I mean, people are making really good videos with this; I just need to figure out how to prompt this a little bit better. This

Viking in the snow is really impressive. I wonder if I can actually see the prompt on it. So if I click on it, we can see a video of a Viking walking through a mountain snowy landscape near the ocean. So you can see the ocean in the

Background and the snowy landscape that's pretty good, with a pretty basic prompt. This one, with the ship on the water, was an anamorphic extreme wide shot of a double-masted wooden ship departing from the ancient port of Joppa. And then let's see if I hover over it.

You can see the whole prompt. I'm not going to read the whole thing; it's a long prompt. So this one actually was a really, really detailed, long prompt. Oh, this pixel art one is really good. What is this? Okay, so I can't read that. I

wonder if I can translate it here. So the translation here is: a cut scene for Minecraft in which the player escapes from a gradually frozen village. That looks really cool. I really like the sort of pixel art look of that video. So I think my problem is I wasn't putting

enough detail in the prompts. I think just slightly more detail will work a little better. Look at this close-up of a monkey—like that looks straight out of a nature documentary! Here. So this was a remix of a different video, making him close his eyes. So people are definitely

getting some amazing videos. I'll just hover here on the screen over some of these videos for a second so you can look at some of the generations people are getting. Here's a few more. And one of the nice things is you can go into the featured section or the recent section,

click into any video that somebody else made and see the exact prompt they used, or remix the prompt and tweak it a little bit. So that's pretty cool as well. So Sora is definitely impressive. So here's a video I found on my computer that I generated back in March of 2023.

It's like a monkey on roller skates; that's what we were getting back in March of 2023. When you compare that to what Sora is generating in December of 2024, we have come a long way. Honestly, I know you can find a lot of issues in

these videos, but just looking at where we were in March 2023 to where we are today in December 2024, AI video has made insane leaps. It does seem like you need to give it a little bit more detail in your prompt. Just giving it a monkey on

roller skates, a wolf howling at the moon, things like that, doesn't seem to be enough to get the motion that I want out of it. Also, again, super bogged down right now. People are claiming on X that it's taking them upwards of an hour to

generate a video right now. I think I've been fairly lucky generating my videos in 15 to 20 minutes right now. Uh, so I would probably hold off, like I mentioned, but again, if we look in the Explore tab and see some of the recent videos people

are getting, some of them are pretty dang impressive; others, not so much. But Sora is here, and this is the worst it's ever going to get. I'm excited to play with it some more. Yeah, let's go! This looks fun. Excited to dive in, test even more, and as

I figure out more tips and tricks, I'll likely make some follow-up videos, and uh, we'll dive in even deeper. Hopefully, you enjoyed this; hopefully, you learned a little bit more about what Sora is capable of, and uh, you get a chance to

Play with it; hopefully, it's not too bogged down for you to actually get in and generate some videos. I'm going to keep going, but because the videos are so long to generate, I'll probably start sharing some of my generations over on X

and Instagram and places like that so you can see what I'm actually getting out of Sora. So if you're not following me on those places, I'll probably be posting stuff there because this video is going to take me multiple days to make if I keep on trying to generate

more videos to put it through its paces. Well, it's as bogged down as it is, but Sora is here! Anybody outside of the UK and Europe can go play with it right now. Sam Altman even mentioned this is like the GPT-1 or the

Dolly of AI video generation from Sora. I can only imagine what we're going to see a year from now. All right, that's it; I'm done rambling. I'll see you in the next video. Bye-bye.

You Can Now Use Sora! Here's How...

AI, No-Code, Tech, Futurism - I'm a tech nerd and talk about tech nerd stuff

Matt Wolfe

OpenAI는 AI 동영상 생성기인 Sora를 출시했습니다.

Sora는 다른 AI 동영상 생성기보다 우위에 있는 것 같습니다.

Sora는 유료 ChatGPT 계정으로 이용할 수 있습니다.

월 $20 요금제는 720p로 50개의 동영상(1000크레딧)을 제공합니다.

월 200달러 요금제는 동영상 500개(10000크레딧), 1080p 해상도 및 추가 기능을 제공합니다.

Sora의 사용 가능 여부는 영국과 EU를 제외한 특정 국가로 제한됩니다.

대시보드에는 최근 동영상, 추천 콘텐츠, 저장된 항목, 라이브러리 섹션이 있습니다.

프롬프트, 스토리보드, 리믹스, 블렌드 및 루프에 필터링 옵션을 사용할 수 있습니다.

설정을 통해 사용자는 프로필 정보, 동영상 게시 및 데이터 사용량을 관리할 수 있습니다.

사용자는 크레딧, 동영상 제한 및 해상도 옵션을 포함한 요금제 세부 정보를 확인할 수 있습니다.

프롬프트 입력 상자를 통해 이미지를 업로드할 수 있으며 다양한 사전 설정을 제공합니다.

해상도, 길이, 동영상 변형 개수를 사용자 지정할 수 있습니다.

초기 테스트에는 '롤러스케이트를 탄 원숭이', '달을 보고 울부짖는 늑대', '타코를 먹는 고양이' 등의 메시지가 포함되었습니다.

트래픽 증가로 인해 일시적인 사용 불가 및 서비스 용량 문제가 발생했습니다.

초기 결과에서는 예상치 못한 오류가 발생하는 등 다양한 품질이 나타났습니다.

롤러스케이트를 탄 원숭이 생성은 낮은 해상도에도 불구하고 인상적이었습니다.

서버 부하로 인해 저해상도에서도 생성 시간이 느렸습니다.

달을 보고 울부짖는 늑대와 타코를 먹는 고양이 세대의 결과는 엇갈렸습니다.

이미지에서 동영상으로 변환하는 테스트에서 처음에는 예상치 못한 오류가 발생했지만 결국에는 동영상이 생성되었습니다.

'수박을 저글링하는 해적'이라는 프롬프트로 다양한 사전 설정('풍선 세계', '스톱 모션', '아카이브 필름 느와르', '골판지 및 종이 공예')을 테스트한 결과 결과가 엇갈렸습니다.

스토리보드 기능을 사용하면 타임라인을 따라 여러 개의 프롬프트가 있는 동영상을 만들 수 있습니다.

'늑대로 변하는 사람'에 대한 스토리보드 테스트 결과 만족스럽지 못한 결과를 얻었습니다.

'절벽 위의 남자가 보름달 아래 늑대로 변신한다'는 직접적인 프롬프트도 제한적인 모핑을 생성했습니다.

긴 생성 시간으로 인해 발표자는 탐색 탭에서 추천 동영상을 선보입니다.

추천 동영상은 세부적인 프롬프트에서 인상적인 결과를 보여주는 Sora의 잠재력을 보여줍니다.

추천 동영상을 분석한 결과, 자세한 안내 메시지가 더 나은 결과를 가져오는 것으로 나타났습니다.

현재 세대와 과거의 시도를 비교하면 AI 동영상 생성에 상당한 진전이 있음을 알 수 있습니다.

발표자는 더 나은 결과를 위해 더 자세한 프롬프트가 필요하며 서버 과부하가 생성 시간에 영향을 미친다고 결론지었습니다.