Shutterstock ImageAI

Shutterstock ImageAI is a text-to-image diffusion model that Shutterstock and Databricks built together. Using Databricks Mosaic AI, ImageAI creates photorealistic images based on trusted data from Shutterstock's own proprietary image library.

The model is trained to create high-resolution images from text prompts, a useful application for content creation and visualization. The ImageAI foundation is built on large language models (LLMs) and diffusion methods to produce high-quality results.

Shutterstock ImageAI takes advantage of the Databricks Mosaic AI's ability to train and deploy large models efficiently. Among the related tools and technologies are DBRX, Mosaic BERT, MPT, Mosaic Diffusion, Composer, LLM Foundry, StreamingDataset and Evaluation Gauntlet. These tools handle tasks like training, fine tuning and evaluating models.

From a technical standpoint, customers can use a variety of open-source models and tools like DBRX and MPT that are commercially available and can be customized. For example, you can pay $20 to pretrain a BERT model on your own data using Mosaic BERT.

Pricing details for Shutterstock ImageAI aren't clear, but the costs of related services and tools are relatively low, so the technology should be within reach of many customers. And Shutterstock's large image library, built into the service, means the images are based on trusted data, which is critical to maintaining image quality and credibility.

Published on June 14, 2024