Google presents Veo 2, an AI model to generate video that will compete with Sora

Google’s strong response to OpenAI’s Sora was to be expected, and it didn’t take long to arrive: Veo 2 is a new AI-based text-based video generator; In addition, it has two advantages over its rival in terms of resolution and duration. The tool can generate content up to 4K and two minutes long, compared to the Full HD limits of the agent incorporated into ChatGPT. At the same time, The Mountain View giant unveiled the curious system called Whisk.

How does I See 2 work?

The announcement comes directly from Google’s experimental DeepMind division, confirming the latest rumors and unveiling Veo 2, the new video generation model based on text prompts. The way it works is still the usual: you write a sentence, as detailed as possible, describing not only the subject or objects, but also the actions, the light, the context and anything else you consider relevant. The system will collect the information and after an indeterminate wait, you will obtain the results.

Here you have an example. The prompt was: “Medium shot, shallow depth of field, a pretty, cartoon-style girl with wavy brown hair, sitting in a 1980s kitchen. She has a small, slightly upturned nose and small, rounded ears. She is expressive.” and gesticulates happily as he speaks to the camera.” Google will apply the invisible SynthID watermark to certify artificial origin and will gradually make the model available through Google Labsstarting with the usual beta testers. It is not yet clear when it will reach the general public.

OpenAI officially launches Canvas, its tool for programmers

The tool that offers real-time corrections and feedback through a canvas is now available to everyone.

The novelty of Whisk

Google also presents improvements to its AI image generator, Image 3, with even more precision and speed of action, and the novelty Whisk: a system again dedicated to images, but not based on textual indications, but on the reworking and remixing of other images proposed by the user.

Article originally published in WIRED Italy. Adapted by Alondra Flores.