AI Has Gone Too Far... This New AI Crossed the Line into Creepy Realism!

If you've been following the AI video scene lately, you know, things are getting pretty crazy. It feels like every other day there's some new breakthrough and honestly, it's hard to keep up. Well, I've got something really exciting to talk about. ByteDance, the company behind TikTok, is making some serious moves in the AI video generation race. The same ByteDance that practically rules the short form video world is, is now charging ahead in the AI space. And it's fascinating to see what they're cooking up. So they've just rolled out two new AI models, Pixel Dance and Seaweed. And these could really shake things up when it comes to generating videos from text prompts. And when a company like bytedance, which already dominates social media, dives into AI video tools, it's a big deal. Alright, so let's break down these two AI models. First up, we've got Pixel Dance, which is still in private beta at the moment. Right now only a select few users can try it out, but. But it could be released to the public next month depending on, get this, the outcome of the US general election. Yeah, you heard that right. According to YouTuber Tim Simmons, who focuses on AI tools, this could go live after November. Apparently there's some political tension around the release, but I'll leave that discussion for another time. So what makes Pixel Dance special? Well, for starters, it focuses on character animation. It can generate 10 second video clips with characters that move super realistically. We're talking about things like walking, turning and interacting with objects in ways that that look so natural it's hard to believe it's AI generated. You can practically see the characters walking through a scene, picking up objects or making gestures that look like they were captured by human actors. But where Pixel Dance really shines is its multi shot capabilities. Most AI video generators struggle to keep visual consistency across different shots. Things fall apart when switching angles or perspectives. This AI model solves that by keeping characters, appearances, proportions and scene details and consistent across multiple shots. Which makes it ideal for creating complex scenes while maintaining visual coherence. A real game changer for AI video production. Not only that, Pixel Dance offers camera control on par with other major models like Pika and Runways Gen 3. You can create impressive camera movements, things like 360 degree pans, zooms and tracking shots with just a text prompt. One of the demos ByteDance showed featured a woman in sunglasses. The prompt was something like in black and white. The camera is shot around the woman in sunglasses, moving from her side to the front, and finally focuses on a close up of the woman's face, and the result was absolutely stunning. It's that level of camera control that puts Pixel Dance in a league of its own. Now let's talk about its sibling, Seaweed. While Pixel Dance focuses on character animation, Seaweed takes on environmental generation. This model can generate longer videos stretching up to 30 seconds or even as much as two minutes, which is pretty rare for AI tools right now. And just like Pixel Dance, it keeps things consistent across shots. This makes it super useful for creating longer scenes or sequences where you need everything to flow together smoothly. The timing of ByteDance's launch couldn't be better either. Everyone have heard about OpenAI's Sora, which was announced back in February. Sora was supposed to be this revolutionary AI model capable of generating up to 60 seconds of high quality video from textbook prompts. People were hyped about it, but Sora still hasn't been released to the public. That's why ByteDance is swooping in to fill that gap with Pixel Dance and Seaweed. Now, even though OpenAI's Sora is still nowhere to be seen, they've been busy working on some new tools that are definitely worth checking out. Just recently, OpenAI introduced a suite of new tools aimed at fast tracking the development of AI voice assistants. These new tools will make it easier for developers to build voice applications using a single set of instructions, which is a big deal for anyone working on voice enabled AI. Previously developers had to go through a multi step process, transcribing audio, generating responses and converting text back to speech. But now with this streamlined system they can do it all in one go. This will save developers a lot of time and simplify the process of creating more advanced responsive AI voice tools. Not only that, OpenAI also unveiled a fine tuning tool that allows developers to further improve their models and by fine tuning them with images and text. This opens up possibilities for better image recognition and more accurate object detection, which could be a game changer for applications in industries like autonomous vehicles. Another intriguing feature is the prompt caching tool, which helps cut development costs by reusing previously processed text, making it more affordable for smaller companies to get in on the action. OpenAI's tools are pushing the envelope on efficiency with which is critical given the growing competition in the AI space. Alright, now there are other big players trying to dominate AI video generation and they're all going hard. Take Kuaishou for example. They launched their model Cling AI back in June and it's already one of the top tools out there. Clingai is integrated into their video editing app and can generate two minute videos just like Seaweed. But Clingai has its limitations. It mostly generates single shot takes meaning it's not as versatile when it comes to complex scenes with multiple angles or camera movements. Still, Cling has racked up over 2.6 million users, and those users have created over 27 million videos already, which is a serious volume. Then there's Pika Labs, one of the OGs in AI video. They just upgraded their tool to Pika 1.5 and this thing is wild. Pika 1.5 comes with more realistic movement, big screenshots, and these crazy special effects they call Pikafects that practically break the laws of physics. Think about characters in your video getting crushed, exploding, or revealing hidden layers like virtual cake. The tool is already live and people are sharing their mind blowing creations on social media. But let's not get too sidetracked. ByteDance's pixel dance and Seaweed models are built on the Dubao family of foundational models, which are based on something called the Document Image Transformer architecture. It's super technical, but essentially this architecture allows ByteDance to optimize these models for things like business applications. This could potentially lower the cost of producing AI generated videos, which is awesome news for anyone who's been scared off by the high prices of AI tools in the past. Speaking of costs, ByteDance's strategy of slashing prices has really paid off. Since May, they've reduced their cost by so much that they've triggered a price war with other Chinese tech giants like Alibaba and Tencent. This aggressive Pricing has fueled ByteDance's growth, and as of now, they're processing over 50 million images and 850,000 hours of speech every day. That's insane. Growth in just a few months. Now let's switch gears a bit and talk about the hardware side of things. ByteDance is planning to develop a new AI model that's trained primarily using Huawei chips. This move comes after the US started restricting the export of advanced AI chips like those from Nvidia, which has been a go to supplier for many AI companies. ByteDance is now leaning on domestic suppliers like Huawei for its AI chip needs. Specifically, they're using the Ascend 910B chip, which is mainly used for less computationally intensive tasks like inference. Basically, when an AI model is already trained and it's making predictions. But here's where things get interesting. ByteDance is planning to use these chips for training new AI models. Training an AI model is a much more demanding task that requires a lot of computational power, so this is a big step for them, but they're facing supply issues. ByteDance has ordered over 100,000 Ascend 910B chips this year, but has only received less than 30,000 as of July. This slow pace is making it hard for them to fully develop their new model. On top of that, the Ascend 910B chips aren't as powerful as Nvidia's GPUs, so ByteDance is still struggling to keep up with the performance levels they need. ByteDance is also one of the largest buyers of Nvidia's H20AI chips, which were specially designed for the Chinese market to comply with U.S. trade restrictions. They've also been buying chips through Microsoft's cloud computing services, making them Microsoft's biggest client for Nvidia chips in Asia. So as we can see, bytedance is definitely positioning itself to be a major player in the AI video generation space. With Pixel Dance and Seaweed, they're tackling some of the biggest challenges in AI video production, like maintaining visual consistency between shots and extending video lengths without sacrificing quality. But at the same time, they're up against some tough competition, and especially from companies like Kweishu, Pika Labs and of course, OpenAI with their yet to be released model Sora. The race is on and it's going to be fascinating to see who comes out on top. Anyway, that's all for this video, and if you're a content creator, these tools could be a game changer for you, so it's worth staying updated on all of this. Thanks for watching and don't forget to like subscribe and drop a comment below about which AI tool you're most excited about. Catch you in the next one.