Зображення Imagen 4 / Google
Google has announced Veo 3, a generative video creation model with audio capabilities, Flow, a movie creation tool, and Imagen 4, which has improved image quality.
According to Google, Veo 3 not only improves the quality of generated content compared to Veo 2, but also «for the first time» creates videos with audio that matches its content. A city street will be able to get traffic noise and crowds, while in the forest you can hear birds singing.
Even conversations are available — at a very good level, as in the examples. This feature includes precise lip synchronization. The company says that it has improved the understanding of user prompts.
Google Veo 3 received not only sound but also several new features compared to the previous version. These include improved work with reference images of scenes, objects, and styles that will help the user get a more accurate result according to their intent. New camera controls help with rotation, zooming, and moving the frame. The «Redraw» function allows you to expand the frame, change the image orientation, and adjust the video to the desired size. You can also add and remove objects.
Veo 3 has been available in the U.S. since yesterday for Gemini subscribers, Ultra subscribers, Vertex AI enterprise and the new Flow app. This tool is designed specifically for use with the Veo. It makes it easy to create cinematic clips with a combination of Veo, Imagen, and Gemini capabilities.
«Use natural language to describe your shots in Flow, manage your story components — cast, locations, objects, and styles — in one convenient place, and use Flow to turn your story into beautiful scenes».
Flow is available for subscribers of Google AI Pro and Ultra plans in the United States and soon in other countries.
The Imagen 4 «image generator combines speed with precision to create stunning» images. Google says «is a huge step up in quality from its predecessor. Improved reproduction of details like animal fur and the use of styles, both realistic and abstract.
The Imagen 4 can capture images in a variety of aspect ratios and resolutions up to 2k. The model is available in Gemini, Whisk, Vertex AI, as well as in Slides, Videos, Docs, and other tools in Workspace.
Source: Google