Google draws its weapons against OpenAI in the generative AI war. DeepMind, the AI research laboratory of the famous Mountain View firm, unveiled Veo 2 on Monday, a new generation of its AI video generator. This new product from the web giant counterattacks Sora, a similar tool that OpenAI made available to the general public last week, promising better results than the competition. Succeeding Veo announced in May, Veo 2 allows you to create videos on demand from a text query. It is capable of generating content reaching a definition of up to 4K (compared to 1080p for Sora) and a duration of up to “several minutes”.
Bringing “a greater understanding of real-world physics and the nuances of human movement and expression,” Veo 2 is capable of imitating cinematic genres and effects. For example, he can take specific shots, such as a low-angle tracking shot, a close-up, or even suggest the use of an 18mm lens or the addition of a shallow depth of field. The examples presented by Google show animated films and other sequences displaying striking realism.
Google highlights the fact that Veo 2 is less prone to “hallucinations,” phenomena that cause unwanted details to appear, like extra fingers or unexpected objects, compared to competing AI models.
Google DeepMind also lifted the veil on Imagen 3, an advanced version of its Imagen image generator, a rival, among others, to OpenAI’s DALL-E. The firm also specifies that it places emphasis on the fight against disinformation, indicating that its image and video generators integrate the SynthID system, which adds an invisible watermark to generated content in order to identify it as being created using ‘AI.
Not officially accessible from Europe, these two new services are respectively available via the VideoFX and ImageFX tools on the Google Labs platform.
Finally, Google also unveiled Whisk, a new AI tool that allows users to create images based on the subject, style and context of other images, rather than relying on textual descriptions.
