On Wednesday, Google unveiled its newest artificial intelligence (AI) models for generating images and videos, marking a significant development in the realm of generative AI that was initially previewed at Google I/O over six months ago. This latest launch is now available on Vertex AI, aimed at enterprise clients. Notably, while Imagen 3 has not been offered as a standalone application until now, it has been incorporated into various platforms, including Google Docs, Gemini, and a testing tool dubbed GenChess.
Google Imagen 3, Veo AI Models
In a blog post, Google announced the arrival of its new AI models on the Vertex AI platform. This managed machine learning (ML) service on Google Cloud enables developers and businesses to construct, implement, and oversee AI models. The platform is comparable to services like Amazon Bedrock and Microsoft Azure, providing integrated solutions for AI-related workflows.
According to the announcement, the Veo video generation model is now accessible on Vertex AI in private preview, allowing organizations to produce videos from either text or image prompts. Meanwhile, Imagen 3 is slated for release next week, enabling enterprises to create images that represent their brand identity and logos based on text prompts.
Veo boasts the ability to produce high-quality videos utilizing text or image stimuli and can deliver a variety of cinematic styles. Developed by DeepMind, this AI model shows strong adherence to prompts and can consistently generate footage of objects and people, realistically capturing their movements.
Imagen 3, available next week on Vertex AI, promises the capability to generate photorealistic images across diverse styles. Described by Google as “our most capable image generation model yet,” this tool can interpret natural language prompts, allowing users to obtain desired results without needing to delve into technical specifics.
The Imagen 3 model will also feature editing tools for inpainting and outpainting. Companies have the option to incorporate their brand colors, styles, logos, and other distinctive elements into the generated images.
Concerning privacy and security, Google has implemented several protective measures. Each image and video frame generated by these AI models will include SynthID, a watermarking technology created by DeepMind, to prevent misuse linked to deepfakes and misinformation. Additionally, Google reassured users that their AI models will not be trained on client data, adhering to Google Cloud’s established data governance and privacy protocols.
