Open AI SORA is Public! | First Impressions & Thoughts

Hello everybody. Welcome back to another video. I know I uploaded twice today, but this video was warranted because right after my first upload, Open AI dropped their third day of ship miss or 12 days of OpenAI, and it was Sora AI video generation. This has been long awaited, as many of you know, since all the way back in February of 2024. So let's talk about it. So first up, like I said, February 15th was when Sora was officially announced as something OpenAI had been working on. It's their first AI video generator. And at the time it was easily the most high quality, most impressive generation for video that we've ever seen come out of the AI space. But that was ever so long ago. Even just looking at my AI video generation playlist here on YouTube, you can see that since Sora's announcement, we have had a plethora of announcements in regards to AI video generators, models such as Cling AI, which was insane and to be honest with you guys is right about there at that same sort of quality we got released to us today, where Minimax Meta has their own AI video generator runways Gen3. A lot has happened in the AI video generation space since Sora was announced. So what exactly is different since that initial Sora announcement and what can we expect from this release? Well, this is, as we said, day three of OpenAI's 12 days. And from OpenAI's demo video here, we can see they've actually built an entire website for Sora. You can share video generations on there, whether they're featured or most recent. You can favorite generations. There's a variety of editing tools that come with this website and it all hooks back up to your OpenAI plan. So whether you have ChatGPT plus or Pro, I think a lot of people were wondering specifically whether or not the quality of Generations had improved in comparison to when we first looked at it back in February. And the answer to that, in my honest opinion, is the quality is pretty much the same. It's still definitely state of the art in comparison to what we have in terms of Chinese video generators, minimax, Clang, the new Tencent Open Source Generator, or even non Chinese ones like Gen 3, VDU and the open source LTX video. They're all pretty close to each other. So yeah, Sora is about as good as the rest of the pack. OpenAI gave open source and the rest of the AI world quite a lot of time to catch up to it. And it looks like OpenAI spent their time optimizing their Sora video generation models. For a turbo mode, A mode that has still a decent high quality, but uses far less computer. Now let's get into the website. It has its own storyboard based editor where you can combine the clips together, which is pretty awesome to see right in that website. But again, they aren't the only website to be doing something like this. LTX Studio comes to mind. Even InVideo AI has something like this. They show off building in story mode with multiple clips at a time. First prompting a beautiful white crane with a yellow tail standing in a creek. And then the crane dipping its head into the water and pulling out a fish. They do some of those classic woolly mammoth generations. And they also show how there are multiple options for things like remixing or how many video generations you want at once. And of course there is credits to go along with that. And again, we'll talk about pricing and the plans in this video. But they also have a feature to change resolution and the ability to remix at different strength levels, similar to video to video that we already have with Gen3 and Runway. Here is the end result of that crane video they demoed. Where you could see the crane, you know, is kind of dipping its head in the water trying to catch a fish. And the splashes aren't super realistic, but overall it's not a bad gen. Again, this is pretty much what you can expect from state of the art video gens these days in the AI space. I gotta say, I do really like their remix feature. You can see they took that woolly mammoth generation and turned them into these crazy huge robots. There's also a transition blend feature where we can blend the robots turning back into the woolly mammoths if we want. Designing UI like this is quite a lot of work and there's multiple ways to blend them. You can mix them, sample them, or even do customized blend curves. So quite a lot of work went into the design of this website and the user interface. And I think that was a part of their plan because they saw midway through obviously how fast open source and other companies caught up to their level of video generation quality. So let's talk about using it right now. It's available on the Sora.com website to both ChatGPT plus and ChatGPT Pro users. Here is my first generation. I've been waiting on this over 30 minutes now and still hasn't come through. Obviously the website is getting absolutely hammered and bombarded, so no surprise there. I'm sure that this will clear up and they'll fix and work these servers out. But I can't even get any generations through. So later in the week I'm going to produce another video that's more of a deep dive and my own personal thoughts in regards to how this model performs in comparison to the other ones. But like I said, it does have the Explore feature. You can see recent generations or feature generations. So we can see what the community is making right now and it gives us a pretty good idea of what the model is capable of. Again, all these generations that I see, they're not bad by any means. They're definitely state of the art level. Here is a particularly good example, I think where we've got, you know, the Golden Gate Bridge and an astronaut is riding a horse across that bridge. But if you look closely, there are some screwy details like the horse is sort of sliding across the floor. And obviously some interesting morphing is going on with the two entities here. Not bad at all. Again, this is a particularly good generation, but it shows you the limitations of the model for sure. What's nice is any of these public generations, since this is kind of like a social media aspect to the website. We can, you know, recut, remix or blend or even create a seamless loop of other people's generations, which is cool. I like that. Under the featured cherry picked generations tab, we can see the best of the best. And there is a lot of good stuff in here. There's no doubt in my mind that Sora is a very decent model. It's right up there with the best. But is it the best of all time? Is it the absolute goat? I don't think so. Let's talk about pricing and plans, because this is where it starts to get very difficult to justify OpenAI's Sora release. If you have the $20 a month ChatGPT plus plan, which in my opinion was already pretty much worth the money beforehand, at least for my use cases. You get 1,000 credits worth of Sora per month, which equates to about 50 generations. You can only do one max concurrent generation with a max duration of only 5 seconds, which is not very much, and you have a max resolution of only 720p. And again, this is all for $20 a month. And by the way, there is no way, at least at this point, to top up your credits. Now if you move on over to the $200 a month ChatGPT Pro plan, which of course includes other features like 01 Pro, which is the first announcement we saw for the 12 days of OpenAI, but it also includes up to 500 fast videos. So 10,000 credits worth of Sora generations and you get unlimited or virtually unlimited relaxed videos in the slow queue. You get up to 1080p resolution, 20 second duration and up to five concurrent generations. And you can also download without the watermark. Also, something that they aren't telling you though in this little overview, which I find pretty interesting, is that you can't upload any media containing people to do either photo or video to video that contains actual people. And apparently false flags are also occurring where, you know, a simple photo of cats for example, is flagged as humans and it won't generate. But here's the thing. This isn't just a safety feature. If you pay the $200 a month, you will get access to the ability to upload people. You don't have to have the watermark. So safety kind of isn't really a concern when you're paying $200 a month, which doesn't really make much sense to me. It seems like the safety is more or less an excuse to try to get you to bump up to that next $200 a month plan. And again, that is a ton of money. It might be worth it for some people, especially if you really like the way that Sora generates videos. But man, it's expensive. I will say I am happy they have unlimited, quote unquote relaxed video generations. That's nice to see. Most expensive plans, even from other video generation websites don't offer anything like that. But now we have to talk about a few other things. This Sora release is not open source at all. No code, no waits, no nothing like that. The only way to use it is on the Sora website right now through one of those two plans. Which also means at the time being right now there is no API for Sora. We could probably expect that sometime next year. I'm sure an API will come out, but we don't have one just yet. So there is no pay as you go option at all for Sora. The other really big stick in the mud here is the fact that there are so many other good options for video generation that it makes it difficult to justify Sora at either of its price points. First of all, we have extremely high quality open source video generators at this point. For example, we have LTX video which is fully open source, weights code and all, and can run on consumer hardware. Granted, it's a lower resolution and it can't do up to 20 seconds I don't believe, but if you have even a pretty decent GPU you can run this completely free on your own machine at home. And again, the quality, as you can see from these little examples here, isn't too far off of what you see from Sora. Plus you can upload, you know, any photos of people that you want and then convert those into videos. And being open source, it's only going to get cheaper and cheaper to run. And any other little improvements made by the open source community can be applied to this model as well. And we're already starting to see that. Not to mention, we also have the Huan Yan video coming from Tencent. And again, this is fully open source with weights, meaning the community can modify it in any way. And this one you can pay as you go on a website like Fall AI, because it can be hosted anywhere it's open source. To do a video on this, it is about 40 cents per video and for a buck you can run it around three times. And just from my first glances, my first looks, this appears to be about the same quality of Sora, if not a little bit better. And to be able to run this three times for a dollar, that puts it around the same price as one month of those Sora plans. And that's for 600 generations. That will be done in about four minutes or so, which seems to be about the time of a fast Sora generation in ideal circumstances. Again, I'm still waiting in the queue for this to even start generating, and it's been well over 30 minutes. The servers are getting hammered right now, so I don't want to judge it too harshly. This was a very, very highly anticipated release. I gotta say, so far community reactions have been really mixed surrounding Sora. A lot of people are just complaining about basic user issues. Again, the servers are getting absolutely hammered. But you see in this tweet, I criticized it for not even being able to do 1080p generation without having to shell out that $200 upfront for the plan. People are not being able to upload even animated characters, which doesn't make sense to me at all if safety is their concern. Issues with signups. Some people are still defending OpenAI though, saying give them time to sort all these day one issues out and fair enough, we'll have to see how it holds up over time. People praising things like the storyboard option, which is genuinely something that you don't see too much in the AI video generation space right now. But again, a feature similar to this can still be found on other sites. And because OpenAI is doing it pretty much every video generation Website is probably going to include some sort of a storyboard feature because they all have to copy the crowd, right? Magospace here points out that at 40 cents a render for like we just saw with Huanyun Video, $200 unlimited runs doesn't sound too bad, but you really have to be generating constantly for that to be worth it. I'd also like to bring this up as well. They put clearly a lot of effort and work into building a nice website to generate with Sora and actually create things and explore your creativity. And I really appreciate that. But I really feel like they missed the mark in regards to the plans. I mean they're directly connected to your existing ChatGPT accounts and I think that's a good thing. I think that's something they should keep. But I also think that folks should have the option to purchase a separate Sora only plan where all of their money can be put towards generating with Sora instead of getting access to those other chatgpt plus or pro features like 01 etc. Etc. I will say though, if this generator can do anything better than other models, it's going to be good diversity. It can do live action stuff, it can do animated stuff, it can do those 3D Pixar style renders. It's really quite diverse. This car rolling, for example here is pretty darn impressive at first glance. We've got a pug chilling with an R in people under some blossoming trees, a woman walking through the desert in a nice cinematic way. People are also clearly testing it with text. Doesn't seem to be able to do great text out of the box, but that's something we can say for most video generation models. Unless you're doing image to video with text generated by an image generator. We've got some nice CCTV footage, again, very diverse in terms of the different styles. It can do nice close up of a cat. We've got this crazy bubble doing all kinds of things. I think to have my final thoughts in regards to Sora, we're going to have to wait until the servers clear up and I can actually start to get some generations through. I've seen a lot of huge W's, like you know this video of a crab doing a cute little hop. We've got some 3D bears dancing. That looks great, but I've also seen some pretty big flops like in this video over here. Obviously this guy is supposed to be eating something, but we can't really tell what it is and he's not really putting it in his mouth. And chewing it, as you'd expect. Or, you know, this first person video that's clearly supposed to be a car driving down the road or maybe a train. But, you know, this perspective is all out of whack. We're going sideways. We've got a weird steering wheel with the gauges going everywhere. Just doesn't make sense. Or this video of a truck that's sort of, you know, morphing and sliding on this highway as it goes down the road. Again, guys, I'd really like to test these features out for myself and really play with SORA before I make any final judgments on it, so to speak. But that's just not possible right now. And I wanted to offer my first initial thoughts on this announcement because I think it's kind of a controversial one. I think there's a lot of mixed opinions and I really want to know what you guys think as well. Is this worth it? Do you think the bump from $20 to $200 is just a little bit too high? I don't know. There's a lot to think about and there's a lot to say. And we also have to give it time to see true comparisons and tests between this new SORA model and the rest of the models that exist. So, yeah, for now, I'd love to hear your thoughts. Thank you. So, watching this video, also check out my other news video that was announced today. There is some really good stuff in there, including unprecedented levels of video generation control, which totally blew my mind. And again is one of those things where open source and some of these smaller community releases can be really, really good and compete directly with giants like OpenAI. See you guys in the next one. Goodbye.