As the year draws to a close, Google has unveiled its new image and video generation models, Imagen 3 and Veo 2, the latter of which could be a serious competitor to OpenAI's Sora.
Google claims to have developed the most powerful video generation model of the moment with Veo 2. It promises groundbreaking performances, including a better understanding of real-world physics, the movement of bodies and objects, and the nuances of human expression. According to Google, the model can adapt to a multitude of cinematographic effects (tight or wide shots, low-angle views, front or rear tracking, etc.) to create unprecedented sequences in ultra-high definition (4K) and up to several minutes long. On paper, this is even more impressive than what OpenAI offers with Sora, a tool that is certainly impressive, but which sometimes lacks fluidity and can also have difficulty coordinating certain movements. Moreover, Sora does not allow the creation of sequences longer than 20 seconds in high resolution (1080p). All videos generated by Veo 2 include a SynthID watermark that is invisible on screen, but which identifies the resulting videos as being generated by artificial intelligence.
Alongside this, Google has unveiled Imagen 3, the latest version of its image generator. It guarantees brighter and generally better-composed images, in a wider variety of artistic styles, from photorealism to impressionism, abstract and animated.
Finally, Google has released Whisk, an experimental tool capable of generating new images without prompts, but using other elements, photos or drawings, as sources of inspiration. By combining several visuals, such as a portrait and a landscape, for example, the tool creates new, unique and personalized images.
Veo 2 and Imagen3 are currently reserved for YouTube creators and companies using Google Labs' VideoFX and ImageFX. Whisk is currently available as an experimental tool, also via Google Labs and only in the United States. – AFP Relaxnews