Luke Jones | Unsplash
There is no doubt that AI is advancing rapidly. But this has led to a series of potential competitors in each of the fields of this science. On one hand we have Sora, from OpenAI, which at one point surprised us greatly. Although we are just now witnessing its shortcomings when seeing a new competitor that is doing everything very well, exposing that Sora still has a fairly green development, making video generation inefficient.
This time, Google DeepMind has released its latest generative AI version regarding video generation. It goes by the name Veo 2 and it seems that its capabilities are far above Sora. The way these videos are created is through the formulation of a prompt, which are nothing more than instructions you must give to the generative model to provide context and specify details. The more precise that instruction is, the closer the result will be to what you want. As a relevant fact, the tool used allows you to include an image to have a starting point, to which the information you enter through the prompt will complement.
Google DeepMind | Veo 2
Another quality is that it is able to create videos in 4K quality, followed by a faithful representation of reality (simulation). These movements are truly realistic and maintain a superior fluidity to what has been seen so far. Considering that these types of parameters are being perfected more and more, we could think that at some point, many of the works we usually see in the film industry might be partly created by generative AI. This could pose a problem for many actors, but if we look at it from another perspective, it could serve in adding special effects, so that a movie can be astonishing because it will have no limits, and the production cost would also decrease considerably.
The same company behind this development admits that there is still much to be done to maintain coherence with what the user requests and to result in scenes that are as accurate as possible. Here again is where digression comes in, or in summarized terms, when AI invents things that have nothing to do with what is requested, resulting in such a disaster that a video can be completely ruined; it's something we saw when this technology began to take its first steps. I think we can remember with a smile that video of Will Smith eating a plate of noodles, but later that same video was recreated with current technology.
Post of Joseph Carlson on X
The interesting thing about Google DeepMind is that it is behind several projects that mostly involve artificial intelligence. Gemini is probably the most representative and is starting to encompass everything (You may have noticed its integration in our Chrome browser!) Probably, these technologies will be perfected over time, and considering the use by millions of people around the world, their training will improve many of today's capabilities.
For now, the things that Veo 2 can create are astounding. For this, the AI analyzes the prompt, trying to delve into each of the details offered, following the context and its coherence. Additionally, it is able to rely on what style it can generate, allowing for a wide range of top-level angles and movements. Physics comes into play, something that the team behind Google is trying to understand in accuracy. I imagine that beyond AI, there is a hard work related to the complete understanding of how physics follows certain laws with absolutely simulated precision, and this is achieved in scenarios of pure research.
This technology is amazing, isn't it? Look at the sample they have created, it is certainly amazing and leaves much to think about the future that awaits us. I hope you enjoyed this article, best regards!
- Main image edited in Canva.
- Translated from Spanish to English with Hive Translator.