[12:39 Wed,15.May 2024 by Rudi Schmidts] |
At the Google IO conference that started yesterday, Alphabet not only presented a variety of new, partly multimodal language models to keep Open AI somewhat in check. An independent AI model for generative video generation was also presented as a sneak peak. Google Veo generates 1080p videos with improved consistency Veo is said to be able to generate high-quality 1080P videos from text, image and video prompts, "capturing the details of instructions in different visual and cinematic styles". This makes it possible to ask specifically for things like aerial or time-lapse shots. Successful videos can also be extended in time and further refined with the help of additional prompts. It should soon be possible to use Veo in Google Labs with an experimental tool called VideoFX - provided you are in the right country and are selected from the waiting list. However, the DACH region still seems to be completely excluded from the sneak peek. Some demo clips can be viewed on a demo page from Deep-Mind. Since it can be assumed that every company only presents the best outputs of its model at such a presentation, it has to be said that Sora&s well-known clip selection has not yet caught up with Google in terms of quality in our eyes. At the same time, the few Veo samples shown so far look much more rounded than anything else Sora has shown in the last 3 months. It must also be emphasized that Sora has not yet been officially released - and therefore it is not possible to estimate how "accurate" it will be to work with either of the two models. more infos at bei deepmind.google deutsche Version dieser Seite: Google DeepMind Veo erzeugt 1080p-Clips mit verbesserter Konsistenz |