After the image, Stability.ai presents Stable Video Diffusion: a generative AI capable of producing video directly from your PC, without going through the cloud.
We talk a lot about ChatGPT, Dall-E or Midjourney, which are very impressive generative AI tools already widely used in the industry. The problem is that they all require enormous computing power in the cloud. This is not the case with Stable Diffusion, which can run locally on your machine to create images, and now videos.
The company Stability AI has, in fact, announced a new open source model: Stable Video Diffusion, its first model capable of generating video.
2 second videos
For the moment, Stable Video Diffusion is limited to two-second videos in 1024 x 576 pixels, which is very limited in the 4K era. More precisely, the generative AI will create 25 images consistent with the user’s request.
This is the whole challenge of this type of model compared to the simple creation of images. To generate a video, you must succeed in generating a succession of images consistent with each other with a logical evolution of a scene.
At the moment the model generates a video from an image, but Stability already promises a version capable of generating a video from a text command.
The strength of the model is that it can run on a personal PC. On Github, we can already read users and get started with Stable Video Diffusion on PCs equipped with a GeForce RTX 4090.