© Luma AI/YouTube
We obviously no longer present AI models like ChatGPT, Claude 3 or Gemini, which excel in text generation. Artificial intelligence is now opening up to new frontiers such as music (with Eleven Labs or Suno for example) or more recently video. OpenAI shook the tech world in February by presenting its Sora AI, capable of generating extremely realistic videos. Well OpenAI is no longer alone in the arena and now has to face Luma AI, a start-up from San Francisco.
The latter has just unveiled Dream Machine, a text-to-videoaccessible via their website. It is possible to try it for free: type the description of the video you want in the generation bar and let the magic happen.
Subscribe to Lemon Squeezer
A new tool accessible to all, but which remains to be perfected
According to the promises of Luma AI , their system is capable of produce high-quality videos in just two minutes after your request. These are then presented in the form of short five-second clips. Early testers reported that Dream Machine was able to very faithfully reproduce the requested characters, objects, actions and environments in a consistent manner.
After several in-house tests, the reality is a little different (see video below). Dream Machine indeed succeeded in generating the requested video, the original prompt of which was: “ A horde of dragons ridden by knights at sunset in a heroic fantasy bucolic landscape”. However, due to high demand, generation took between three and seven minutes and the result is very far from the presentation trailer that you can see above.
The character modeling still remains quite shaky and we can see that the dragon on the left is nothing like a dragon. The 3D itself remains quite poor, the lighting is correct, but the overall atmosphere of the video is… strange. Namely: the video you see here is the most convincing that we managed to obtain after three tries.
After having tried other prompts by varying the themes and graphic styles, we can affirm: Dream Machine does not’ ;is clearly not comfortable with photorealism.
Let's temper it all the same: Dream Machine remains a new tool and will surely be improved over time. The fact that it is so easily accessible and already offers more or less convincing renderings still remains quite impressive. But the proof is that you should never trust beautiful presentations.
A desire for democratization
Where his rivals, such as Sora d&# 8217;OpenAI, or Kling, from Kuaishou, remain confined to the restricted circles of a few partners, Luma AI has chosen to take a decidedly more inclusive path. By making its Dream Machine open to all, the company clearly seeks to encourage the creativity of as many people as possible, which cannot be taken away.
The company also plans to create APIs and plugins for creative software. This deliberately more open approach could allow Luma AI to quickly build a vibrant community of creators and developers gravitating around its platform.
While the result is not yet perfect, Luma AI is making a fairly notable debut with Dream Machine. It's also worth remembering that AI video generation is still in its infancy, and even the most advanced systems can produce strange and inconsistent results. The choice to make such a powerful tool so easily affordable must be underlined, even if we can also question the potential abuses of its use.
- Luma AI has just unveiled its video generative AI, Dream Machine.
- Easily accessible via their website, it allows you to generate short five-second videos from a simple prompt.
- Currently, Dream Machine productions are not yet perfect and remain still far from the company's promises.
📍 To not miss any news from Presse-citron, follow us on Google News and WhatsApp.
[ ]