April 10 news.Googlefor itsVideo AI modeling Veo 2 Introducing new features designed to help users more easily generate cinematic-quality footage as well as edit real video. These added Veo 2 features are now available via Google Cloud's Vertex AI platform for preview, alongside updates to Google's text-to-image generator Imagen 3 and audio-related AI models.
According to 1AI.New features in Veo 2 include inpainting and outpainting.Google says the Repair feature automatically removes "unwanted background images, logos, or distracting elements" from the video, while the Expand feature expands the original video using AI. Google says Repair automatically removes "unwanted background images, logos, or distracting elements" from a video, while Expansion expands the frame of the original video by filling in the added space with AI-generated video footage that naturally blends in with the original, similar to the generative expansion feature that Adobe offers for images. The tool fills in the added space with AI-generated video footage to naturally blend in with the original footage, similar to the generative expansion Adobe provides for images.
The update also allows Veo 2 users to select movie shooting technique presets along with text descriptions when generating footage.in order to guide shot composition, camera angles and pacing in the final result. For example, presets include time-lapse effects, drone perspectives, and simulated camera pans in different directions.
In addition, Veo 2 has a new interpolation function.Ability to create video transitions between two still images, fills new frames for the beginning and ending sequences.
Adobe's competing Firefly video model has some similar features, and its generative AI video extensions launched last week in Premiere Pro. Google has also added the SynthID digital attribution watermark to its AI-generated output, similar to Adobe's content credentialing system. However, Adobe further promises that its tools are fully commercially available because they are trained on licensed and public domain content, a standard that Google, which heavily crawls the web when training its AI models, is unable to meet.
In terms of the text-to-image model Imagen 3, Google has updated its editing features, claiming to "dramatically" improve the automatic removal of objects so that results look more natural after distracting elements have been removed. Veo 2 and Imagen 3 are already being used by companies including L'Oréal and Kraft Heinz for their marketing content production. Production. Justin Thomas, head of digital experience at Kraft Heinz, says that what used to take eight weeks to accomplish now takes eight hours.
In the audio space, Google has launched a private preview of its text-to-music model Lyria and an "instant custom speech" feature for its synthetic speech model Chirp 3. Chirp 3 can now generate "realistic customized speech" from 10 seconds of audio input, according to Google, and it's launching a preview of a new transcription feature that recognizes and separates out individual speakers in a multi-participant conversation to provide a clearer transcription of a multi-participant conversation.
These updates are just a few of a number of AI-related announcements Google made today. The latest version of Google's efficiency-optimized Flash model, Gemini 2.5 Flash, is coming to Vertex AI, which Google says "automatically adjusts processing time" based on the complexity of the task to provide faster results for simple requests.
Additionally, Google is updating its Agentic AI tool for the enterprise this week, enabling AI agents to communicate with each other and perform tasks across different platforms, such as PayPal and Salesforce. Meanwhile, the Google Cloud Marketplace will launch a new section for organizations to browse and purchase AI agents built by third-party Google partners.