Google is broadening its artificial intelligence toolkit for business clients, revealing substantial updates to its media generation models during its Cloud Next conference on April 9, 2025, primarily targeting its Vertex AI platform. Demonstrating quick follow-through, the company’s Veo 2 video generation model started appearing for select users on April 11, accessible through Google’s AI Studio platform via the Gemini API.
Veo 2 Video Tool Emerges On AI Studio
The initial rollout gives some US-based users hands-on experience with Veo 2. Early access appears limited, and users outside the US might need technical workarounds like VPNs to even see the option. Current capabilities, as reported by early testers and confirmed in Google’s developer blog, are set at 720p resolution and 24 frames per second, with a maximum video duration of 8 seconds.
This access comes at a cost, pegged at $0.35 per second of generated footage, a factor likely to influence its adoption for various projects. This initial deployment follows the Cloud Next announcement where Google detailed new Veo 2 features entering preview on Vertex AI, including enhanced editing capabilities and camera controls for effects such as background removal or time-lapses. The separation between the feature preview on Vertex AI and the model rollout on AI Studio suggests different access pathways for enterprise developers versus individual creators or testers.
Lyria Adds AI Music Generation
Alongside video, Google introduced Lyria, its text-to-music AI model now in private preview on Vertex AI for approved customers via an allowlist request form. Google suggests Lyria can produce complete musical pieces across different genres from text instructions, potentially aiding businesses in creating custom audio for marketing campaigns or immersive experiences, potentially reducing reliance on traditional stock music libraries and their associated licensing complexities.
Voice Cloning And Smarter Imaging Join The Mix
Audio capabilities also saw development with Chirp 3, Google’s audio generation and understanding model, becoming generally available on Vertex AI. Following an earlier preview phase reported around March 2025, Chirp 3 now formally includes an ‘Instant Custom Voice’ feature, which Google claims can clone a voice from just 10 seconds of audio.
This feature supports HD voices across 31 languages with multiple speaker options and operates under a stated verification process to check for proper usage permissions. Chirp 3 also introduces a transcription tool designed to distinguish between different speakers in recordings.
Image generation wasn’t left out. Imagen 3, which saw an earlier preview rollout on Vertex AI, received further enhancements announced at Cloud Next. These improvements focus on better reconstructing missing image parts (inpainting) and higher quality object removal, aiming for more realistic and detailed image output compared to previous versions.
Google Sharpens AI Edge For Business Users
These updates across multiple media types underscore Google’s push to make Vertex AI a comprehensive platform for enterprise AI development, particularly in media creation. Google’s official blog highlighted Vertex AI as uniquely offering generative models spanning video, image, speech, and music. This strategy places Google in direct competition with other cloud AI platforms like Amazon Bedrock. The Cloud Next event also saw broader AI announcements, including new custom silicon like the Ironwood TPU and frameworks such as the Agent Development Kit (ADK), signaling a wide-ranging AI focus for Google Cloud customers.
With the increased capabilities, especially in areas like voice cloning and media synthesis, come questions about responsible use. Google stated that media generated by its newer models like Imagen, Veo, and Lyria incorporate SynthID watermarks. The company also referred to built-in safeguards against harmful content creation and the specific verification process for the voice cloning tool.
However, Google did not specify the datasets used for training these powerful models. This remains a sensitive topic industry-wide, with ongoing legal challenges concerning the use of copyrighted works for AI training. Google has pointed to its opt-out mechanisms and an indemnity policy aimed at protecting Vertex AI customers from related copyright disputes.
