View attachment 34536
If Veo 2 is able to develop a gymnast that moves like a human, then it will be able to beat Sora.
It is possible that Google has only recently started distributing its Veo generative AI to enterprise customers; yet, the company is not wasting any time in releasing a fresh version of the video tool to early testers. During the week of Monday, Google revealed a sneak peek of Veo 2. A statement made by the firm asserts that Veo 2 "understands the language of cinematography." As a practical matter, this indicates that you are able to make reference to a certain type of film, cinematic effect, or lens when you are urging the model.Furthermore, according to Google, the new model has a better knowledge of the mechanics that exist in the real world as well as individual movement. Creating an accurate representation of humans in motion is a challenge that every generative model needs to overcome. Consequently, it is noteworthy that the firm asserts that Veo 2 is superior in terms of both of those problematic aspects. It goes without saying that the examples that the business has provided are not sufficient to know for certain; the real test of Veo 2's capabilities will be when someone instructs it to make a film of a gymnast performing their routine from scratch. To add insult to injury, Google has said that Veo will produce artifacts such as additional fingers "less frequently." This is in reference to the challenges that video models face.
View attachment 34537
The media is reporting that Google is working on enhancing and refining the app for Imagen 3 in an entirely separate way. The company claims that the latest version of their text-to-image technology is better at producing more lifelike images with more balanced compositions. Furthermore, it can render a wider spectrum of artistic styles with better accuracy than before. And, it even excels in complying with prompts much better. This was a concern I raised about prompt adherence when Google rolled out Imagen 3 to customers of Google Cloud earlier this month. At the very least, this testament goes to show that Google understands where it needs to improve in terms of its AI models.
Gradual roll-out to users in the United States will begin to see Veo 2 come on stream through Google Labs. For the time being, however, Google hopes that these inner testers will restrict video capture to a maximum of eight seconds at 720p resolution, which in short means that Sora can record no more than twenty seconds of footage at 1080p, but to do this, a $200 per month ChatGPT Pro membership must be purchased. And all users of Google Labs from more than a hundred countries are also provided access to the current updates of Imagen 3 through ImageFX.
