In recent years, we’ve heard more and more about the possibility of generating images with artificial intelligence, through text prompts: brief descriptions of what we hope to generate, which an AI transforms into an image. These tools have improved significantly over time, but In the case of AI-generated videos, progress is still limited.
Google wants to intervene in this direction, and for this it presented Lumiere, an artificial intelligence model capable of creating videos using text commandsand the results promise to be much smoother than we have seen so far in other proposals and alternatives.
Lumiere can not only create videos from text, but also edit existing videos
What makes Lumiere more interesting is that, unlike other companies’ developments in text video generation, Lumiere does not require joining images to create a video, but rather You can create the video in one go, so the result is much smoother and natural, avoiding creating videos with temporal inconsistencies. This is achieved thanks to a new system developed by its creators, which they have named Space-Time U-Net.
Lumière officials explain that they used “a dataset consisting of approximately 30 million videos including text as description” to train their artificial intelligence model, without offering more details, at least until now.
All this translates in to several abilities for Light
- Create videos from text– You write the description and Lumière creates it
- Create videos from images: Which means you can convert still images into animated videos.
- Generate stylish image-based videos: The AI can use an image as a style reference when creating videos requested by the user.
- Edit videos– You can edit parts of an existing video. For example, in a video of a woman running, it is possible to ask “that the woman is made of flowers”, and the AI will succeed.
- Making “cinemagraphs”: or in other words, animate specific parts of images
What are the (current) limits of Light?
At the time of publishing their study, the creators of Lumière explain that their AI can generate up to 80 images to create videos up to 5 seconds long and a resolution of 1024 x 1024 pixels.
And that’s it Light is still developingor, and for this reason, it is not yet accessible to the public. It is unclear when we will be able to test this AI and when Google will be able to implement its technology in one of its final products. After all, the company has big ambitions when it comes to artificial intelligence, with the arrival of Gemini and the integration of various AI-based features directly into the Android operating system.
Lumière may not be ready yet, but it is still an AI that is already offering promising results. The question is: how realistic will your videos be in 2, 3 or 5 years?
In Xataka Android | If Google Photos’ magic eraser left me speechless, the Pixel 8 Pro trick is pure witchcraft
