Google upgrades Vertex AI with new generative tools

Google has announced updates to its generative AI models on the Vertex AI cloud platform.

These updates include enhancements to text-to-music, video creation, voice cloning, and image generation tools.

Lyria and Veo 2 are now in preview, enabling music creation in various styles and enhanced video editing tools like background removal and time-lapse.

The Chirp 3 audio model is generally available, supporting voice cloning in 35 languages and speaker separation in multi-person audio recordings.

Google’s Imagen 3 image generator has been updated to improve object removal and image repair capabilities.

All media generated by Imagen, Veo, and Lyria is watermarked with Google’s SynthID technology for transparency.

🔗 Source: TechCrunch

🧠 Food for thought

1️⃣ Enterprise AI platform competition intensifies as a three-way race

Google’s updates to Vertex AI models like Lyria, Chirp 3, and Imagen 3 represent strategic moves in an accelerating enterprise AI platform competition.

Market data reveals a significant gap between competitors, with Amazon Bedrock currently leading with 3,411 customers (0.69% market share) compared to Vertex AI’s 740 customers (0.15% market share)1.

Each platform has developed distinct competitive advantages. Google Vertex AI leverages deep integration with Google Cloud and analytics capabilities, while Amazon Bedrock offers more foundation models and stronger AWS service integration2.

The specialized media generation capabilities (music, voice, video) that Google is emphasizing suggest the competition is moving beyond basic text generation into more specialized creative domains where enterprise customers seek proprietary solutions.

2️⃣ Content creation platforms face unresolved copyright challenges

Google’s silence about its training data sources reflects a widespread industry challenge, as legal battles continue over whether training AI on copyrighted works constitutes fair use.

Recent court decisions have produced conflicting precedents, with a federal court ruling against Ross Intelligence for training on Thomson Reuters’ materials3, while other cases remain pending with arguments centered on transformative use.