In one year, generative AIs have evolved from “simple” text assistants, to image creators, to video generators, as demonstrated by Stability AI’s latest invention: Stability Video Diffusion.
If you follow the news of major language models and AI at all, the name of Stable AI should be familiar to you. Although this company is less known than OpenAI or Google, it has already made AI such as Stable Diffusion which, like Midjourney or DALL-E, allows you to generate images from text descriptionsor even Stable Doodle, which transforms your sketches into drawings.
The company’s engineers relied on their previous research to present today the next evolution of their artificial intelligence: the creation of more or less realistic videos in high resolution. The latent video broadcast model created by Stable Diffusion makes it possible to generate images and videos from text descriptions. Their style can range from realistic photography to abstract art. To achieve this result, Stability AI added a temporal dimension to AI creations and “adjusted each frame to small, high-quality video datasets.”
This AI allows you to create high-quality videos from your text descriptions
That said, if you want to make a feature film with Stable Video Diffusion, don’t rush to Github right away. Current technology currently only allows the generation of 2 seconds at 25 frames/s maximum (SVD being offered in two forms, SVD and SVD-XT) at a resolution of 1024 x 576 pixels. It doesn’t matter, because the company insists on this point: “this model is not not intended for real-world or commercial applications at this stage. Your opinion and feedback on safety and quality are important to refine this model for possible release. »
Despite these limitations, watching the video published on YouTube will be enough to convince you: with SVD, Stability AI is indeed opening the era of AI-assisted video creation for everyone.