This is a big deal.
As you know, earlier this year we saw Sora AI and, recently, a new Chinese AI video generator that is also nearly as good as Sora AI. But we didn't have access to those. Dream Machine by Luma Labs, however, is currently free to try on their website. And that is a huge deal. Let's check this thing out!
An Awesome Video Generator
Like I said, folks, Luma Labs is not messing around. Dream Machine is a high-quality video generator. This interactive website is pretty cool; we can hover over these videos to slow them down and watch them. There are realistic-looking people and characters that look like they're straight out of an animated movie. It's amazing, and it's especially awesome because we can access it right now.
More About Dream Machine
Dream Machine is an AI model that produces high-quality, realistic videos fairly quickly, with text and images as possible inputs. Therefore, you can animate your Midjourney or Dolly 2 generations using this model. They note that it is highly scalable, which means that the more GPUs we have, the higher-quality generations we can get. It doesn't cap out like previous AI video generators. They also describe it as efficient; it's a transformer model trained directly on videos. So the generations are physically accurate, consistent, and eventful. They say it's eventful because older video generators like PABS or Runway ML were slow-moving without much realistic action. Dream Machine is Luma Lab's first step towards building a "universal imagination engine.".
Amazing Quality for Free
And it's available to everyone right now for free, which is crazy! The examples are breathtaking. We're finally seeing these high-quality Sora competitors come to the public, and this is the first one we have access to. You can put in your prompts, and it does around 1 frame per second of generation. The shots have lots of smooth, realistic motion—people riding vehicles, characters with props, and scenes straight out of movies. It can do different styles too, not just realism but also cartoons, video games, etc. The quality and consistency are really impressive stuff—it's competitive with Sora, for sure.
Current Limitations
The current limitations include morphing, movement, text, and genres. But still, this was a shocking free release from Luma Labs.
Trying It Out
So I tried a typical prompt: "closeup of an anthropomorphic lemon sipping lemonade on the beach.". The server seemed overloaded at first, and I couldn't get generations through. But people on my Discord server got some going.
The first one of a guy eating a lemon is quite horrifying with his mouth! But the movement is so impressive. We also got a more realistic version of him eating the lemon. And one of a guy eating spaghetti—difficult prompts, but it did a pretty good job considering.
More Awesome Examples
There are lots of creative, nuanced prompts people are trying that weren't possible before without huge effort. Like a retro TV with butterflies or a cinematic girl overlooking another world. The reflections, aspect ratios, and different styles, such as anime, look great.
Luma Labs says video-to-video input will be coming soon too, where you can input a video and edit it, giving yourself crazy hair or superpowers! From kids doing that in iMovie to now... just awesome creative potential.
There are examples on Discord of scenes from video games, Call of Duty-style shots, and even horrifying but impressive deepfakes of me eating a burger! The movement and details, like consistent UIs in games, show its outstanding capability.
Current Free Usage and Cost
Right now, there is a restrictive cap of 30 free generations per month. After that, it gets very expensive: $30 for 120 more, $100 for 400, and up to $500 for 2000 per month. So expect people to try using many free accounts!
I'll be doing a live stream soon when their servers calm down, taking prompts from you all to further explore this amazing new technology. The creative potential of AI-generated video is just mind-blowing. While not as excellent as the private Sora yet, having any access at all to this quality level is insane.
Conclusion
This is just the start—the worst this tech will ever be. As it improves further, the ability for one person to create an animated movie or video game by simply prompting an AI is incredible. Let me know your thoughts!
0 Comments