Databricks & Shutterstock launch enterprise AI image model

Databricks and Shutterstock have announced the launch of Shutterstock ImageAI, a text-to-image generative AI model designed for enterprise use.

The model, powered by Databricks, utilises the capabilities of Databricks Mosaic AI and is trained exclusively on Shutterstock’s extensive image repository. The resulting AI is able to generate high-fidelity, bespoke images tailored to specific business requirements.

Traditionally, organisations in need of high-quality images for their various platforms had limited choices. They could either select from a repository of licensable assets or opt for costly production efforts. With the advent of AI-generated images, new possibilities emerged; however, these models often fall short in meeting the high standards required by enterprises due to less rigorous training data and a lack of integration with enterprise applications, impacting data governance and observability.

ImageAI aims to address these gaps by providing companies with a tool to rapidly and confidently create new, high-resolution, photorealistic images suitable for commercial use. The model's training on Shutterstock’s curated image collection ensures the reliability and quality of the generated images. Developed on the Databricks Data Intelligence Platform with Mosaic AI tooling, the model leverages the security, reliability and speed of Mosaic AI Model Training. Shutterstock and Databricks were able to train the model from scratch within weeks. Furthermore, Databricks clients can deploy ImageAI using Mosaic AI Model Serving, allowing seamless integration with their existing enterprise-wide data governance frameworks.

“At Databricks, we believe that companies can and should be in the driver’s seat when it comes to building their own custom GenAI models on their data. Shutterstock offers one of the most comprehensive visual datasets in the world, and we're thrilled they chose to build their production-quality, text-to-image model on top of the Databricks Data Intelligence Platform,” said Naveen Rao, Vice President of AI at Databricks. “We already have trained thousands of custom models for our customers, and we expect many more leaders like Shutterstock to build models powered by Databricks that they can provide as a service.”

Aimee Egan, Chief Enterprise Officer at Shutterstock, commented on the collaboration, stating, "Shutterstock is excited to partner with Databricks to bring ImageAI to life. This collaboration underscores our commitment to advancing responsible AI and providing our customers with innovative tools that enhance their creative workflows. Whether for marketing, design or product needs, ImageAI creates trusted visuals that businesses can own. It also offers unparalleled customisation and safety for enterprise use, and customers can be protected by Shutterstock indemnification. It represents a significant step forward in the integration of AI in content creation."

The primary benefits of Shutterstock ImageAI include the ability to produce trusted outputs. Organisations can generate images that adhere to their requirements with the assurance that the model was trained on Shutterstock's proprietary and vetted image repository. Shutterstock clients also have the option of receiving indemnification on outputs for added protection.

The model is adaptable to unique business needs, enabling organisations to generate new high-quality images swiftly. Future updates will allow businesses to fine-tune the model with their corporate brand identity using Databricks Mosaic AI Model Training to produce images unique to their company's aesthetic.

Secure integration with enterprise applications is another key advantage. ImageAI can be accessed through Databricks, allowing enterprises to incorporate the model as an API call into their applications. This integration comes with automatic governance and monitoring capabilities.