Amazon Web Services (AWS) has introduced the Amazon Titan Image Generator, an AI-powered tool designed to create images from text descriptions or modify existing ones, following the lead of Microsoft and Adobe. Unveiled at the AWS ‘re:Invent 2023’ conference, the generator allows content creators to efficiently enhance images using English language cues.
It supports industries like advertising, e-commerce, and media and entertainment by enabling the creation of high-volume, cost-effective, studio-quality images. The model is capable of understanding complex instructions involving multiple elements, producing contextually relevant images.
Trained on diverse, high-quality data, the model aims for accuracy, generating realistic images while promoting inclusivity and minimizing distortions. AWS highlights the customization potential, allowing companies to fine-tune the model with their proprietary data to align with brand guidelines or create images in a specific style.
The Titan Image Generator includes responsible AI usage features, such as an invisible watermark on all generated images to discreetly identify AI-generated content and curb misinformation spread. AWS also introduced two new Amazon Titan multimodal foundation models along with Amazon Titan Multimodal Embeddings.
The models encompass 25 years of AI and machine learning innovation at Amazon, providing high-performing options for image, multimodal, and text models through a fully managed API.
Additionally, AWS announced the integration of Anthropic’s Claude 2.1 foundation model in Amazon Bedrock, enhancing accuracy and reducing hallucination rates, especially in long documents and system prompts.
The AWS Generative AI Innovation Center has collaborated with global customers since its June 2023 launch, assisting in conceptualizing, prioritizing, and constructing tailored solutions leveraging generative AI.
AWS also unveiled the AWS Generative AI Innovation Center Custom Model Programme for Anthropic Claude, allowing customers to fine-tune models securely using their proprietary data starting Q1 2024.
