Google introduces Veo 2, a stunning artificial intelligence video maker + video

Google unveiled Veo artificial intelligence 7 months ago and now Velo 2 It has arrived. Veo 2 can produce 2-minute clips with a maximum resolution of 4K (4096 x 2160 pixels); In other words, compared to Sora AI, Google’s new model offers videos with 4 times the resolution and about 6 times more time.

According to Google, the Veo 2 model can produce 2-minute videos with 4K quality. Of course, currently you cannot make videos with such quality and these figures are only theoretical. In Google’s experimental video production tool, VideoFX, where the Veo 2 is now exclusively available, videos are limited to 720p and 8 seconds. Its competitor, the Sora, can produce 20-second 1080p clips.

Google also updated the text-to-image model Image 3 introduced, but its changes are not so much that it can be considered as a completely new model and just an improved version.

Google’s Veo 2 video generator AI

If you want to use the Veo 2 model, you need to go to the VideoFX platform and wait on the waiting list; However, Google says it will increase the number of users who can access the new AI this week. Veo 2 will also be available through the Vertex AI developer platform, according to Google. Overall, the company wants to bring the new model to different parts of its ecosystem in the coming months, based on user feedback.

Like any other video generator model, Veo 2 also creates videos for you through text prompts, but this model has some advantages. Google’s new model can produce clips in a wide range of styles, and its understanding of physics has improved and its videos are “clearer”.

Sharper means that textures and images in clips are sharper, especially in scenes with a lot of movement. Also, the control of the camera has been improved and in Veo 2, the virtual camera can be better placed in different angles.

Google claims the Veo 2 can more realistically model the dynamic movement of fluids (such as pouring coffee into a glass) and light properties (such as shadows and reflections). Google claims that this model suffers less illusions and does not produce the common errors of other models such as extra fingers, but this claim cannot be accepted with certainty.

Google has not yet announced what the Veo 2’s training resources were, but YouTube videos are one possible source. Google owns YouTube and has previously said that YouTube content “may” be used in the development of models such as Veo.