Google apparently wants to stand up to OpenAI and Sora. The Google Deepmind team has released a new version of the video AI tool Veo. In the first tests, users are enthusiastic – even if there are still problems.
Only recently did OpenAI make its video AI Sora accessible to users. But Google apparently doesn’t want to let this move go without a backlash. Just a few days after OpenAI’s announcement, the Google Deepmind team is showing a new version of the video AI Veo.
Veo 2: What the AI can do and how it compares
Recommended editorial content
Here you can find external content from X Corp.which complement our editorial offering on t3n.de. By clicking on “Inhalte anzeigen” You agree that we may provide you with content from now and in the future X Corp. may display on our pages. Personal data may be transmitted to third-party platforms.
Note on data protection
As Google revealed in an announcement, Veo 2 should be able to generate minute-long clips in 4K resolutions. In addition, the video AI should better represent movements compared to its predecessor – be it of people, animals or objects or the movements of the “camera” that records the scene. This should offer filmmakers many options for their To have the scene generated exactly as you imagine it.
Google has also created its own comparison tests between Veo 2 and competing video AI tools. As always, these should be taken with some caution, as Google – and other AI companies – have not advertised results that put their products in a bad light. In the surveys, Veo 2 is said to have been preferred by participants in two respects: the quality of the video output and the accuracy of the prompt implementation.
The competitors in the test included Meta Movie Gen, Kling v1.5 as well as Minimax AI and Sora Turbo. According to Google, Veo 2 is said to have performed particularly well compared to OpenAI’s video AI. Here, almost 67 percent of participants preferred the output of Veo 2 and almost as many said that the Google AI implemented the prompt better.
Veo 2 still has limitations
On X and other platforms, users have already given the AI a number of tasks. For example, users have found that Veo 2 can actually generate better physics representations compared to Sora and other video AIs. This includes, for example, cutting a tomato or people being able to eat spaghetti without accidents.
Recommended editorial content
Here you can find external content from X Corp.which complement our editorial offering on t3n.de. By clicking on “Inhalte anzeigen” You agree that we may provide you with content from now and in the future X Corp. may display on our pages. Personal data may be transmitted to third-party platforms.
Note on data protection
However, Google itself shows in a few examples that Veo 2 is not yet free of errors. There are problems with consistency within scenes if they are longer. For example, people in the background can suddenly disappear, merge or appear out of nowhere. There are still challenges when it comes to physics, as the angler’s clip above shows. However, Google wants to improve problems like rubber fishing rods in the future based on feedback from Veo users.
Access to Veo 2 is currently regulated via a waiting list. Anyone who wants to get on the waiting list in this country is unfortunately looking down the pipes. Clicking on the button in the announcement leads to a page that says: “VideoFX is not yet available in your country.” In the future, Google wants to activate even more users for Veo 2.
As Techchrunch reports, even users who have access still face restrictions. Currently only clips with a maximum resolution of 720p and a length of eight seconds can be generated. Here Sora is ahead for the time being with a resolution of 1080p and 20 seconds.
These are the best clips from OpenAI’s video AI Sora
View image gallery
Don’t miss any news about software & development
Note on newsletter & data protection
