Will you soon be able to create your own movie in seconds, with just a few words? After having revolutionized image creation (with software like Midjourney and Dall-E) and then text writing (with ChatGPT and its little brother GPT-4), artificial intelligence is attacking a new field: video creation.
US startup Runway unveiled a new version of its “Text to Video” tool on Monday. It allows you to generate video sequences in a few minutes, simply by typing in a few words of English the scene you want to appear.
This type of AI is still in its infancy: Runway’s AI can only render sequences of about 3 seconds, with moving edges and sometimes inconsistent content. And this feature is not yet accessible to everyone, as you have to sign up for a waiting list.
But like their text counterparts, these models move at an impressive speed. The first version of Text to Video, released by the startup in February, only allowed editing the style of an existing video using AI.
This new version retains this feature and adds others, such as the possibility of combining a description with an image that will serve to guide the style of the generated video. The videos currently created are silent, but Runway is also interested in generating audio, its manager told Bloomberg.
A booming sector
Runway isn’t the first company to work on this type of AI. As early as fall 2022, tech giants Meta and Google had announced that they were working on their own video-generating AIs. But neither has announced when those shows will be available to the general public.
The video sector is also explored by the current champion of generative artificial intelligence, OpenAI, as revealed by its leader Sam Altman. But despite the speculations, their latest GPT-4 model is not capable of such creations.
Another “text-to-video” model has been released to fill the vacant space: an AI soberly called “text-to-video synthesis”, created by a research lab affiliated with the Chinese giant Alibaba, is available from March 19 on the as a free demo on Hugging Face, one of the essential platforms of the artificial intelligence boom. It also promises to create short video sequences from just a few words.
But this demo is regularly saturated due to the large number of users. And, above all, the videos generated by this AI regularly include the logo of the Shutterstock platform, which suggests that the AI has been trained on a large number of videos on this platform, raising doubts in terms of copyright.
These models are also bound to raise many ethical questions. The more photorealistic videos AIs can generate, the more accessible deepfake creation becomes, even for disinformation purposes or pornographic deepfakes.
Could these shows soon revolutionize the hitherto largely preserved cinema? seizures that are already affecting many artists? Some industry professionals have spoken out against these developments, such as Steven Spielberg, who said “very uncomfortable” with AI creations: “The soul is unimaginable and inexpressible. It cannot be created by any algorithm.”
Source: BFM TV
