Veo 2, the successor to DeepMind’s original Veo model, can create videos with a resolution of up to 4K and a duration of over two minutes. It currently leads the race with OpenAI’s Sora system, which can produce 1080p videos of up to 20 seconds.
Improved realism and display fidelity
The key improvements of the Veo 2 system are described by the DeepMind team as follows: “Veo 2 donosi poboljšani realizam i vjernost prikaza, značajno nadmašujući druge AI video modele u pogledu detalja, realističnosti i smanjenja neželjenih artefakata. Model se ističe naprednim mogućnostima prikazivanja pokreta zahvaljujući razumijevanju fizike i sposobnosti praćenja detaljnih uputa. Također nudi veću kontrolu nad opcijama kamere, precizno interpretirajući upute za stvaranje širokog raspona stilova snimanja, kutova i pokreta, kao i njihovih kombinacija.”
Imitations of different types of lenses
As they say, the capabilities of the model go beyond the technical specifications. Veo 2 shows a more advanced understanding of cinematography, which includes the ability to emulate different types of lenses and cinematic effects, such as shallow depth of field, which could be especially useful for creative professionals.
It will be widely available “kada model bude spreman”
Currently, access to the Veo 2 system is limited through a waiting list. The model is available through Google’s experimental video creation tool, VideoFX, where videos are currently limited to 720p resolution and eight seconds in length. Eli Collins, vice president of products at DeepMind, told TechCrunch that Google will provide access to the Veo 2 system through its Vertex AI development platform. “kada model bude spreman za korištenje u širem opsegu.”
The field of AI video generation is increasingly competitive
The introduction of Veo 2 comes at a time when the field of AI video generation is becoming increasingly competitive. In addition to OpenAI’s Sora, other notable systems include Runway’s Gen-2, Pika Labs’ Pika 2.0, Stability AI’s Stable Video Diffusion, and Meta’s Make-A-Video. Kling, Minimax (also known as Hailuo), Luma Labs’ Dream Machine, and Haiper AI are also making progress in this area, each with their own strengths.
Google DeepMind, like other companies in this field, implements safeguards such as content filters and watermarking technology to reduce the risks associated with videos generated by artificial intelligence. This variety of AI video generation tools indicates a growing demand for such technology in a variety of industries, from entertainment and marketing to education and professional video production.