A few days before Christmas, Google DeepMindthe AI branch of Googlehas unveiled a new tool that is likely to get people talking. You knew Sora with OpenAIget ready to discover Veo 2the latest innovation from the creators of Gemini. But on the French side, you will simply have to look from afar, because the product has not yet been made available for France, even in a trial version.
“ Make the results more realistic ”
As the name suggests Veo 2 is the second version of the artificial intelligence model capable of generating videos of Google. In the space of less than a year, the international firm has succeeded in improving its first generative AI model for incredible results.
Indeed, if we are to believe the press release, the latest tool of Google reportedly rated as one of the best generative tools on the market today. “In head-to-head comparisons performed by human reviewers, Veo 2 achieved industry-leading results against leading models” can we read in the document of Google.
The objective, with Veo 2 is to counter one of the major problems encountered by generative AI models: hallucinations. Indeed, it sometimes happens, even if our prompt is almost perfect, that the tool misinterprets something or generates something that was not requested.
Most of the time, this kind of detail is found on the hand side when we want to generate a human, where it is not uncommon to see a sixth finger appear in the human anatomy. According to Googlethis problem would have been corrected, even if errors still occur.
“Veo 2 produces them less frequently [ndr, les hallucinations]which makes the results more realistic” we read in the press release. And the results are stunning.
Translation : “Veo 2 is capable of:
- Create videos in resolutions up to 4K
- Understand camera controls in prompts, such as wide shots, POVs, and drone shots
- Better recreate real-world physics and realistic human expression […]”
Google does not only unveil a video generation tool
If Veo 2 is truly the flagship product put forward by Google, the firm behind the world's most famous search engine also presented an update to its image generation model: Imagen 3.
Translation : “We've also improved Imagen 3's ability to:
- Produce a variety of artistic styles: realism, fantasy, portraiture and more
- Transform prompts more faithfully into precise images
- Generate clearer, more compositionally balanced visuals”
The annoying question: are we going too far?
This debate will inevitably take place, perhaps even in the public space or before that, at the Christmas meal. Is image and video generation going too far? Or are we perhaps going too fast?
These are questions that can be asked at a time when it is becoming increasingly easy to tamper with or generate images to make them say or do what we want. During the American presidential period, we were entitled to numerous deepfakes broadcast on social networks and if some were really crude, others were really very well done.
If we persist in wanting to give all the keys to realism to AI as with Google And Veo 2will we always be able to find the true from the false? And this is just the tip of the iceberg.
What about artists and intellectual property? What about the resources used to run the servers?
Thanks to ever-increasing competition in the sector, the progress of AI is undeniably astonishing and each time pushes the limits of technology a little further, more and more quickly and in all sectors.
But perhaps we need to know how to take the time and not push certain limits that we would not be able to recross once they are exceeded.
Source : DeepMind
With an unwavering passion for local news, Christopher leads our editorial team with integrity and dedication. With over 20 years’ experience, he is the backbone of Wouldsayso, ensuring that we stay true to our mission to inform.
