Google Cloud Enhances AI Media Tools with New Music and Video Features

On Wednesday, Google unveiled upgrades to its Vertex AI cloud platform, boosting several proprietary media-generating AI models.

Lyria, Google’s text-to-music AI, is now in preview for select users, while the Veo 2 video model gains advanced editing and visual effects options. Additionally, Google introduced a voice-cloning feature powered by Chirp 3, its audio processing model, for approved users. The Imagen 3 image generator now offers notably improved performance, according to Google.

Announced at Cloud Next, these updates strengthen Google’s bid to lead the enterprise generative AI market, competing closely with Amazon’s Bedrock platform, which provides similar AI tools.

Google markets Lyria as a replacement for royalty-free music libraries, enabling users to produce songs in diverse styles, from jazz piano solos to lo-fi beats.

Chirp 3 supports speech synthesis in approximately 35 languages. Initially previewed earlier this year, it powers Instant Custom Voice, which can replicate a voice using just 10 seconds of audio and is now widely available. Chirp 3 also enables a new preview tool, Transcription with Diarization, which distinguishes and labels speakers in multi-person recordings.

To curb misuse, Instant Custom Voice undergoes a verification process to ensure proper voice usage permissions, Google says.

Veo 2 now allows users to remove backgrounds, logos, or objects from videos and extend video frames, such as converting landscape footage to portrait. It also supports adjusting camera angles and pacing for AI-generated scenes, creating effects like time lapses or drone-style shots, and can interpolate between specified start and end frames.

These Veo 2 features are currently in preview.

Google noted that Imagen 3’s enhancements improve its ability to remove objects and restore missing or damaged image areas.

All media created by Imagen, Veo, and Lyria (but not Chirp) carries Google’s SynthID watermark. Google emphasized that its generative AI models include safeguards to prevent harmful content creation.

Google did not disclose the training data for its models, a topic often debated due to intellectual property concerns. Some companies train AI on copyrighted material without permission, claiming U.S. fair use protections, though creators frequently challenge this in court.

Google previously informed TechCrunch that it provides opt-out options for model training and an indemnity policy to protect Google Cloud and Vertex AI users from AI-related copyright claims.