Amazon is launching an image maker — joining the ranks of many, many other tech giants and startups that have already done so.
During a keynote at the AWS re:Invent 2023 conference this morning, Amazon introduced the Titan Image Generator, which is now available in preview for AWS customers on Bedrock, Amazon’s AI development platform. As part of Amazon’s Titan family of AI models, the Titan Image Generator can generate new images (with text description) or adapt existing images.
“[You] can use the model to easily swap out an existing one [image] background with a tropical forest background [for example]Swami Sivasubramanian, VP of Data Services and Machine Learning at AWS, said on stage. “[And you] can use the model to seamlessly swap backgrounds to create lifestyle images while keeping the main theme of the image and creating a few more options.”
Amazon says the Titan Image Generator was trained on a “diverse set of datasets” across a “wide range of domains” and can optionally be fine-tuned on custom datasets and includes built-in mitigations for toxicity and bias. (Barring the tests, the jury is out on how effective those mitigations are, of course.) The company declined to say exactly where those data sets came from — and whether it licensed from or reimbursed all the creators of the images used for the Titan Image Generator training.
Image Credits: AWS
Some companies that develop image generators, such as Stability AI and — as of a few months ago — OpenAI, allow creators to opt out of training datasets if they choose. Others, including Adobe and Getty Images, institute compensation systems for creators — though not always well-paid or transparent.
However, Sivasubramanian claimed on stage that Amazon will protect customers accused of copyright infringement with images generated by the Titan Image Generator — under the AI’s indemnification policy. This is certainly music to the ears of AWS customers worried about regression or when a production model spits out a copy of a training example.
In a recent overview of Fortune 500 companies by Acrolinx, nearly a third said intellectual property was their biggest concern about using genetic AI. Other voting found that nine out of 10 developers “consider a lot” of IP protection when making decisions about whether to use genetic AI.
Images generated using the AWS Titan Image Generator. Image Credits: AWS
“We choose carefully how we train our models and the data we use to do it,” he said. “With these investments, our customers will have the flexibility to choose the best models for their requirements, even as their needs grow and change.”
Images created with the Titan Image Generator will also come by default with an invisible “tamper-resistant” watermark – an effort to mitigate the spread of misinformation and abuse of AI-generated images, says Sivasubramanian. (Deepfakes since the Gaza war and AI generated child abuse images It’s the latest illustration of how big the threat has become.) It’s unclear exactly what kind of watermarking technique Amazon is using, and what tools beyond Amazon’s API will be able to detect it. we’ve reached out to Amazon for clarification.
Sivasubramanian noted that the watermarks are part of the voluntary commitment around artificial intelligence that Amazon signed with the White House in July. Other signatories, including Google and Microsoft, have adopted the Content Credentials system developed by the Coalition for Content Origin and Authenticity or have developed their own solutions.
