It seems like only a few days ago that Google released Veo version 1…. because it was literally a few days ago! The company released its video generation model as a limited preview just before OpenAI released Sora. Well, Google just announced Veo 2.
The age of AI videos is dawning, as companies are both releasing video generation models and streaming services with AI content. The latter example refers to TCLtv+ and the AI videos that will exist on it. While AI videos mostly exist on social media, it looks like it’s going to expand into more corners of the entertainment industry.
After months of waiting, OpenAI finally released Sora to its users. If you’re on ChatGPT Plus or higher, you’ll be able to use the model. Right now, users need to deal with a ton of restrictions. For example, if you have ChatGPT Plus, your video clips can only be five seconds long at 720p resolution.
Google announced Veo 2
Back when Google unveiled Veo 1 in May, it already looked realistic. Well, as with any generational jump, the results we’re seeing in Veo 2 look much better. In fact, they look damn near flawless. Google showed off some short clips, and they are virtually indistinguishable from something recorded with a camera.
We don’t know if these results were one-in-a-million results selected after countless uncanny failures, or if the model is just that good. The results are much better than any other results we’ve seen before. AI videos, even the best ones, usually have some sort of tell letting you know that they were spat out of a model.
However, the results in Veo show the most consistent backgrounds, characters, water physics, shadows, hair, and other finer details. It’s rather scary to think about, as filmmakers are getting more obsolete by the second.
There were only two examples that had some sort of tell. Firstly, the little girl generated in a CGI style never fully closed her mouth. Also, we could see some weird artifacts typical of AI videos around her mouth.
Lastly, the video with the flamingo looked EXTREMELY realistic. The water physics, reflections, and everything else looked spot on. The only thing was that its legs temporarily swapped places; its left leg became its right leg for a few seconds.
Other than that, it looks like Google has won the AI video generation game. The company is releasing this model to VideoFX users. The company will expand it to more people as time goes on.