Google Cloud has officially launched Imagen 2, a substantial enhancement to its image-generation capabilities, exclusively available to Vertex AI customers on the allowlist. Developed with Google DeepMind technology, Imagen 2 offers improved image quality and advanced features.

Users can leverage intuitive tooling, fully managed infrastructure, and built-in privacy and safety features to customize and deploy Imagen 2. Notable features include generating high-quality, photorealistic images from natural language prompts, rendering text in multiple languages, creating logos, and enabling visual question and answer.

Crucially, Vertex AI's indemnification commitment now covers Imagen 2, providing a two-pronged copyright indemnification approach. This addresses concerns and instills confidence in customers utilizing generative AI products. Google emphasizes Imagen 2's enterprise-grade reliability and governance, allowing organizations to create images aligned with specific brand requirements.

Impressions on Imagen 2

Snap is leveraging Imagen for its Snapchat+ subscribers, introducing an AI Camera Mode that enables users to express creativity effortlessly. With a simple tap, users can generate scenes by typing prompts, selecting pre-set options, sharing them with friends, or adding to their stories.

Josh Siegel, Senior Director of Product at Snap, praises Imagen's scalability, safety, and image quality, allowing Snap to focus on design and user experience. Shutterstock, known for ethically sourced AI image generation, enhances its capabilities with Imagen on Vertex AI. The Shutterstock AI image generator turns text prompts into unique visuals, offering over 16,000 searchable Imagen pictures for licensing.

Read Also: X's Ad Revenue Projections for 2023 Signal Major Setback Under Elon Musk

Canva also embraces Imagen on Vertex AI, integrating it into its design process for millions of images generated to date. Danny Wu, Head of AI at Canva, highlights the partnership with Google Cloud, emphasizing how Imagen empowers Canva's community to enhance content creation at scale.

"With Imagen, our 170 million+ monthly users can benefit from the image quality improvements to uplevel their content creation at scale. The new model and features will further empower our community to turn their ideas into real images with as little friction as possible," Wu said, as quoted in the tech giant's blog.

Moreover, Google noted that it performed "robust safety testing" before the unveiling of the technology to prevent the creation of harmful materials. "From the outset, we invested in training data safety for Imagen 2, and added technical guardrails to limit problematic outputs like violent, offensive, or sexually explicit content."

Possible Intellectual Property Issues

Despite these advancements, Google remains silent on the training data used for Imagen 2, sparking legal questions regarding AI vendors commercializing models trained on potentially copyrighted data. This cautious stance contrasts with Google's earlier disclosure of using the public LAION dataset for the first-gen Imagen, which contained problematic content, according to TechCrunch.

Intellectual property protection concerns persist, as Google and rivals like Amazon lack mechanisms for creators to opt out of training datasets or receive compensation. Instead, Google relies on an expanded indemnification policy for eligible Vertex AI customers, guarding against copyright claims linked to Google's use of training data and Imagen 2 outputs.

Recognizing the issue of regurgitation, where AI models produce mirrored versions of training examples, Google acknowledges its validity for corporate users and developers. The expanded policy aims to allay concerns by focusing on indemnification against copyright claims.

Related Article: AI Excels at Imitation But Not in Innovation, Study Finds