OpenAI’s advanced text-to-image model, DALL-E 3 has now been integrated into Microsoft’s Bing browser, making it free to use.
Microsoft had already begun rolling out the text-to-image model on Bing over the last week or so, though it was initially available only to Bing Enterprise users. The feature was later expanded to Bing Image Creator and eventually to Bing Chat.
The most interesting part, however, is that Bing got access to the latest DALL-E model even before OpenAI’s own AI model, ChatGPT. The text-to-image model is scheduled to be integrated into ChatGPT this month, though only paying users would be able to access it.
What’s New About DALL-E 3?
OpenAI claims that DALL-E 3 is a huge improvement over previous models of the text-to-image generator, capable of understanding prompts far better. The new version of DALL-E is more creative and generates more photorealistic images, the company added.
It comes integrated with a host of new safety tools, which also apply to the DALL-E 3 integration on Bing.
For instance, DALL-E 3 cannot be used to recreate images of public figures or create NSFW or hateful images. This is a critical move to help prevent the use of AI-generated and morphed images in spreading misinformation and disinformation.
Additionally, Bing is also getting a content moderation system that aims to remove harmful and inappropriate images.
We have trained our system to follow our terms of service and community guidelines and to avoid images that contain nudity, violence, hate speech, or illegal activities.Microsoft
In its efforts to combat the risk of disinformation posed by faked images, Microsoft joined hands with Amazon, Anthropic, Google, Inflection, Meta, and OpenAI in July to develop watermarking techniques for the detection and labeling of AI-generated pictures.
The latest DALL-E model also uses ChatGPT to tailor and tweak the prompts provided by the users to generate better results.
Such a feature is now in place in the Bing Image Creator, applying an invisible watermark to images created on the browser using DALL-E 3. In addition to marking the image as AI-generated, the watermark also includes the date and time when it was created.
When using DALL-E 3 via Bing or ChatGPT, users can also refine the results by conversing with the chatbot rather than having to try and perfect the initial prompt endlessly.
In June, Microsoft also integrated Bing AI with its all-new Multimodal Visual Search feature, which enables users to include images in their prompts. This advanced AI feature can even execute prompts like recognizing the objects in images and answering questions about them.
Potential Threats Associated With Free DALL-E 3Access
Thanks to its ability to create more photorealistic images than the previous models, DALL-E 3 is a potent tool in the hands of those intending to spread misinformation through faked images. Now available for free via Bing, the powerful text-to-image model has become available to everyone – increasing the risks.
While the new content moderation and watermarking systems will help combat the threat, they might not be 100% effective. One user even managed to successfully trick Bing AI into reading a Captcha by overlaying its image over the picture of a necklace.
“We have large teams working to address these and similar issues”, Microsoft assured, talking about the challenges posed by text-to-image technology.
