핫 샷: 새로운 제너레이티브 AI 비디오 플랫폼 개요

### 핫 샷: 텍스트-비디오 AI 세대의 새로운 경쟁자
* 🎥 핫샷은 소라와 매우 유사한 새로운 제너레이티브 AI 동영상 플랫폼으로, **4명으로 구성된 팀이 개발**했다고 보도 자료에서 강조하고 있습니다.
* 📽 플랫폼의 초기 모델인 Hot Shot XL은 불과 13개월 전 **초당 8프레임으로 1초 클립**을 제작했지만, 이후 더욱 사실적인 동영상 콘텐츠를 생성할 수 있도록 발전했습니다.
* 🖼 사용자는 플랫폼을 무료로 사용해 볼 수 있으며 **하루에 두 개의 동영상을 생성**할 수 있고, 유료 티어는 **대기 시간 없이 무제한으로 액세스**할 수 있습니다.
* 👀 데모 동영상은 **"머리띠를 한 여성이 물 위를 들여다보는 모습"**과 같은 프롬프트가 고품질의 결과물을 만들어내는 인상적인 텍스트-비디오 기능을 선보입니다.
* 🤖 홈페이지의 인상적인 예시에도 불구하고, 개인적으로 동영상을 제작해 본 결과 **팔다리가 사라지거나 비현실적인 애니메이션**이 나오는 등 여러 가지 문제가 발생했습니다.
* 💡 이 플랫폼은 **720p 해상도로 최대 10초 길이의 동영상을 생성**할 수 있으며, 크리에이터가 내년에 **AI가 생성한 전체 YouTube 동영상을 제작**할 수 있도록 하는 것을 목표로 하고 있습니다.
* 📊 향후 발전 가능성은 매우 높지만, 현재 **무료 모델의 한계**로 인해 일관된 고품질의 결과물을 원하는 사용자들은 실망할 수 있습니다.
* 📬 크리에이터는 시청자가 더 나은 결과를 위한 팁을 공유하도록 장려하고, **더 많은 AI 혁신을 탐구하는 데 관심이 있는 사람들을 위해 뉴스레터를 통해 사용할 수 있는 AI 도구 모음**을 강조합니다.

핫샷은 유망한 제너레이티브 AI 동영상 플랫폼이지만, 그 결과는 매우 다양합니다.

while we're still waiting for Sora, we got another generative AI video platform. This new one is called Hot Shot, and you can find it over at hotshot. In fact, this one is so close to Sora that their press release was literally titled How a Four-Person Team Built Sora. They put together a pretty in-depth article here, going into how this model was built, explaining that their very first model, Hot Shot XL, generated 1-second clips at 8 frames per second, and that was only 13 months ago. And now they're getting video quality like this. And here's the coolest thing: anybody can try this out right now. However, before we test this out, let's take a peek at some of the demo videos that they've put out already, showing what it's capable of, starting with the one that we can see on the screen right now, which used the prompt a woman with braids peeking above water, a man with an afro sitting on a cloud, aerial view of a couple on a picnic blanket, a woman

looking worried in front of a refrigerator. We could see on the website they have some categories where we can select different videos and take a peek. So let's look at some others here. A kid with a determined expression. A person looking concerned. A person smiling. Now, as far as text-to-video generation goes, this is some of the most realistic video I've seen. I've seen better image-to-video generation, but for text-to-video, this is pretty dang good. Here's a vintage camera. A time-lapse of a hand painting a bonsai tree. A barista creates art. Snowcapped mountain peaks reflected in a lake. Erupting geyser in a geothermal area. A fly-through between mountains in the ocean. Dense rainforest canopy with sunbeams. A blonde woman on her phone at the subway. A woman looking out a train window at the mountains. A flight attendant walking down an airplane aisle. Here's a tsunami coming through an alley. I mean, that looks pretty dang realistic. A ball of flames in a market. Wisps of smoke. Come on, that

looks pretty cool. News anchor talking directly at the camera. A focused man with a headset making conversation that is ultra-ultra realistic. There might be a little funkiness in the teeth and maybe the microphone looks a little off, but other than that, the person himself looks really good in my opinion. A woman looking scared at the dentist office. It's cool 'cause you can just see the expression in the eyes without even seeing her mouth. Corgi sitting in a school classroom. Anthropomorphic cat taking a selfie in Paris. I mean, the cat kind of had is like three human-looking fingers, but hey, that's AI video for you. Golden Retriever puppies cooking in the kitchen. There are some pretty good Golden Retriever puppies, honestly. A bear in a grocery store holding up a very good sign. Yeah, as soon as they start adding text and signs in there, it definitely looks a little bit janky. Again, sunny side up eggs, sourdough toast, and tomatoes. A jug pouring milk into a glass. Anthropomorphic cat as a butcher cutting salmon. Hey, this time it didn't try to

give it human hand. I mean, it doesn't look like cat paws either, but it didn't give it human hands at least. And those are all the examples they have on their website, probably fairly Cherry Picked. I mean, that's kind of what they do, but they do give us the option to make your own. So let's go ahead and test it real quick ourselves. I'll click make your own, and here's the deal. Right now, it's going to let you generate two images per day for free. So you can see up here, it says due to high demand, we're temporarily limiting three tiers to two generations per day. Paid tiers have no weight times or daily limits. I haven't used either of my two for today. I wanted to wait till I record this video, so you're going to see what I generate with me. I'm going to turn to my old buddy Claude here and have it suggest some ideas for videos for me. I'm going to give it the prompt, Give me a couple interesting prompts to test a text-to-video model. Each prompt

should be less than 10 words and not too complex. The only reason I put that was because when I was showing off the demo prompts that they showed us, none of them were super long, super detailed prompts. They were all, you know, 10 words or less. Let's see what recommends. Number one is a cat riding a skateboard through a neon city. I'm actually going to tweak that a little bit. If you've watched my past AI videos, I always like to do a monkey on roller skates. So let's do a monkey on roller skates in a neon city. I'll keep the neon city from Claude's suggestion just so Claude doesn't get offended. And let's generate and see how it does. I have no idea how long these generations take, but when this is done, I'll let you know how long it took. So I'm actually starting to get some static now, so I'm not sure if that's just like a static video. It's playing in the background or if I'm actually seeing the

diffusion process in real time. All right, so that took about a minute and a half, and here is the monkey on roller skates video that we got. Not great. The arms disappear a little bit, and it looks like it's more of a skateboard-looking thing than roller skates. They look connected. The output's not very good. One thing that was very cool, though, was the process as it was making this video. It was showing the diffusion happening, so you saw those grainy pixels all start to sort of form into a video until we saw the final output, which is the video that you're seeing on the screen right now. So now I only have one prompt left to try, so let's go ahead and use Cla's other suggestion: blooming flowers emerging from melting snow in time lapse. Let's go ahead and submit that prompt and see what it gives us. I'm going to speed up the video this time so you can actually see it diffuse into the video that we're going for because it does look really cool to see it actually start to form.

the video for you, and here's our video. This one's actually a little bit more impressive than the last one. It doesn't really look like a time lapse to me, but I mean, the colors and the flowers and the snow all look pretty dang good. Wonder what would happen if I clicked regenerate. Will it regenerate for me even though I'm out of credits? Nope, it wants me to upgrade. So, it looks like typically you can generate up to 200 Generations per month for 99 bucks a month, or you can do up to 60 Generations per month, but only two per day. I'm going to take one for the team here and upgrade so I can generate some more videos and see what this is capable of. But with the free model, hopefully you get what you're looking for within two generations. I'd say you have to be pretty dang lucky for that to happen, though. I had Claude generate a handful more ideas. Let's go ahead and see what these look like. Here's a chameleon.

changing colors on a rainbow background. I see the chameleon. I don't really feel like it's changing colors though. Here's a sand castle rising and crumbling with the incoming tide. I'm not really seeing the rising and crumbling but I'm seeing the sand castle and the tide. This one's a paper airplane soaring through diverse landscapes. Yeah, this one's just cursed. Sorry. This one is campfire sparks swirling into constellations at night. Okay, so this one is a robot learning to paint a masterpiece, which honestly is probably the best one I've managed to generate so far, and that's not saying a whole lot if I'm being honest. I haven't really been impressed with any of the generations that I've generated. The stuff that they had on their homepage, their examples, clearly cherry-picked because I haven't gotten anything close to that. However, most of the stuff that I was most impressed by on the homepage actually had people in them. It seemed like it did really good with faces and realistic people, so I'm going to try a

couple more prompts this time. I'm going to be sure to include people in the prompts and see how those do. All right, so this one's a ballerina pirouetting in a field of floating bubbles. And well, yeah, her leg disappeared. Appear, she has three arms at one point. Both of her arms go to one side of her body at another point. Yeah, not great. Let's try another person though. And now we have a chef juggling vegetables in a busy kitchen. Now the guy himself actually looks pretty decent, but obviously you could see the vegetables just look like they're floating around and disappearing and acting all funky. It's not a bad generation honestly. I actually think the guy looks good, but yeah, you can see what's going on with his hands. Doesn't look that great. And here's the last one that I'm going to show off is an astronaut planting a flag on a purple alien planet. The planet and the astronaut don't look too bad. The flag is just kind of floating there and there's nothing that says that he's

planting the flag. Now, the day I'm recording this is actually the day that this hot shot tool was released to the public. And since it came out, I've been seeing some generations on X that are actually pretty decent, better than any of the results I've managed to get. Here's one from Heather Cooper: a living tattoo on a person's skin that glitches and animates, telling stories through shifting pixelated imagery that spreads, retracts across the back, cinematic realism, dramatic lighting. Maybe I just need more detail, I don't know. Here's one from Akash: a giraffe walking through a mall with a cell phone or something. That looks pretty dang good. Keep in mind that Akash is somebody that actually works at Hot Shot, so they have the ability to kind of generate as many as they want and cherry-pick the best ones, if we're being honest here. Here's one from Caitlyn Holloway, and you can see she put a pretty basic prompt: chickens walking under the Eiffel Tower.

and that one looks pretty dang solid. Here's another one from Akash here that says Hot Shot can do Styles too, like comic book Styles, but they didn't share their prompt, so I don't know what prompt was used to get this comic book style. Here's another one of like a Mario character walking ground that actually looks pretty good. I think it's pretty funny. I can see that as being like a little meme in a video or something. Here's one from Dream Starter of somebody walking through a field of wheat or something like that. And finally, here's another one from Akash here of somebody putting out a fire in a kitchen. That doesn't look too bad either. Again, when I look at the prompts that they demo on their homepage, they're all pretty dang basic short prompts to get really good videos. But whenever I tried these types of similar prompts, I got videos nowhere near this. So either A, they're really cherry-picking the videos, or B, I haven't figured out the best way to prompt this to get the best results.

yet, so if somebody's watching this video and they have a tip or a trick to get the best results on this style of video, please leave it in the comments. Help me out, help other viewers of this video out. Let's figure out together how to actually get the best possible videos out of this program because so far none of the videos I made came anywhere close to the quality that I'm seeing on the homepage, which is kind of a bummer when you only get two a day, and if both of those kind of stink, you got to wait till tomorrow to try again or pay a 100 bucks. Neither are ideal for most people. All right, before I wrap this video up, there's a couple little things I want to point out. This hot shot tool can generate up to 10 seconds of footage, although I don't think any of the videos I generated were actually 10 seconds. Most of them seem to be about 5ish seconds. It generates them at 720p, but the most interesting thing to me that

they say in this article is that in the next 12 months entire YouTube videos will be AI-generated by creators. Control over every aspect of generations, longer durations, higher resolutions, real-time interactivity, and more modalities like audio are just around the corner. The videos they showed were amazing; the videos I tried to make, not quite as amazing. Where they're headed, really, really exciting. So, that's Hotshot. It's over at Hotshot. Not a sponsored video; they don't know I'm making this video. I came across this today, I saw the news release, and went, Let's go test it. I'm sure a lot of people don't want to go spend the 100 bucks a month. I'll go spend the 100 bucks once and see what I can get out of it and show off what people can expect, because I'm not going to cherry-pick. I'm going to show you exactly what I get out of it, so hopefully, you found this helpful. That's Hotshot, as always. If you want to discover even more cool AI tools, check

out future tools. This is where I collect all the coolest AI tools that I come across. I've also got a free newsletter, so I'll send you some of the cooler tools straight to your inbox. And if you like videos like this, you want to see AI tutorials, AI news, you want to see me test out cool AI products, like this video, and subscribe to this channel, and I'll make sure more videos like this show up in your YouTube feed. Thank you so much for watching, thank you so much for nerding out with me. I really appreciate you. I'll see you in the next video. Bye-bye.

NEW AI Video Tool Claims To Be As Good As Sora!

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

Matt Wolfe

핫 샷: 새로운 제너레이티브 AI 비디오 플랫폼

새로운 제너레이티브 AI 동영상 플랫폼인 Hot Shot은 1초 클립을 생성하는 것에서 불과 13개월 만에 사실적인 동영상 제작으로 빠르게 발전해 왔습니다.

이 플랫폼은 다양한 데모 동영상을 제공하며, 인상적인 텍스트-투-비디오 생성 기능을 선보입니다.

사용자는 웹사이트에서 Hot Shot의 기능을 사용해보고 다양한 동영상 카테고리를 탐색할 수 있습니다.

이 웹사이트는 하루에 두 개의 무료 이미지를 생성할 수 있는 옵션을 제공하며, 유료 티어에는 제한이 없습니다.

사용자는 동영상 아이디어를 묻는 메시지를 통해 모델을 테스트하여 네온 도시에서 롤러스케이트를 타는 원숭이를 만들었습니다.

동영상 생성에는 약 1분 30초가 소요되지만 팔이 사라지고 스케이트가 스케이트보드처럼 보이는 등 결과물이 좋지 않음.

동영상 생성 도구는 생생한 색상, 꽃, 눈 등 인상적인 비주얼을 제공하지만 무료 모델로는 성공에 한계가 있습니다.

유료 요금제로 업그레이드하면 한 달에 더 많은 세대를 생성할 수 있습니다.

이 도구는 특정 개념에 어려움을 겪지만 사실적인 얼굴과 사람을 생성하는 데는 가능성을 보여줍니다.

거품이 가득한 들판의 발레리나는 팔다리가 일그러지고 팔이 떠 있습니다.

바쁜 주방에서 야채를 저글링하는 요리사는 외모는 괜찮지만 야채가 펑키해 보입니다.

보라색 외계 행성의 우주 비행사는 깃발을 꽂기 위해 고군분투하고, X의 다른 우주인들은 핫샷 도구로 인상적인 세대를 선보이고 있습니다.

핫샷은 만화책 애니메이션을 포함한 다양한 동영상 스타일을 제공하지만 이러한 스타일에 사용되는 프롬프트는 공유되지 않습니다.

도구의 데모 프롬프트는 기본적이지만 사용자를 위해 생성된 동영상은 홈페이지에서 볼 수 있는 화질과 일치하지 않습니다.

이 도구는 720p로 최대 10초 분량의 동영상을 생성할 수 있지만, 대부분의 사용자 동영상은 5초 내외였습니다.

AI가 생성하는 YouTube 동영상의 미래

향후 12개월 내에 YouTube 동영상은 완전히 AI로 제작되어 제작자가 길이, 해상도, 상호작용 등 다양한 측면을 제어할 수 있게 될 것으로 예상됩니다.

필자는 동영상 제작을 위한 AI 도구인 Hotshot을 테스트하고 그 결과를 공유할 계획입니다.

또한 멋진 AI 도구에 대한 뉴스레터를 제공하고 시청자가 더 많은 AI 관련 콘텐츠를 좋아하고 구독하도록 장려합니다.