OpenAI teased us with a sneak peek at DALL-E 3 a few weeks ago but now anyone can use the AI image generator for free on Microsoft’s Bing.
In the initial press release, OpenAI said that DALL-E 3 would be integrated into the paid version of ChatGPT. The demo video showing how it would work was impressive but ChatGPT users are still waiting their turn.
Microsoft is also rolling out its DALLE-3 powered Paint Cocreator tool which is a creative assistant in its Paint app.
Once the announcement was made, Microsoft’s servers were quickly overwhelmed by the amount of users wanting to try the new version of DALL-E. CEO of Advertising and Web Services at Microsoft, Mikhail Parakhin, tweeted “We expected some strong interest, but we didn’t expect THAT much.”
Folks, we know DALL-E 3.0 generation right now is taking longer than normal. We expected some strong interest, but we didn’t expect THAT much, especially given it’s a weekend. Bringing more GPUs in, will be better soon.
— Mikhail Parakhin (@MParakhin) October 1, 2023
The promised additional servers must have done the trick because when I tried it out the images were generated pretty quickly.
Microsoft reiterated OpenAI’s claims that DALL-E 3 was a breakthrough in text-to-image generation. The upgraded tool promises more accurate prompt following, more coherence, and improved photorealism and aesthetics.
OpenAI previously hinted at a digital watermark being in the works and Microsoft’s blog post confirmed that it adds an invisible digital watermark that adheres to the C2PA specification. It will be interesting to see if this watermark can be broken as all the others have.
DALL-E 3 has strong content moderation built in so you won’t be able to generate any NSFW images.
The images I managed to generate looked pretty good, albeit not quite up to my expectations of photorealism.
Prompt: a boy and a girl splashing through puddles after the rain, photorealistic
One of the really impressive features of DALLE-3 is how good it is at generating text, which is a common challenge for AI image generators.
Prompt: over the shoulder shot of an old man reading a copy of Tom Sawyer
The images are generated at a resolution of 1024×1024 which is great for web use. Hopefully, they’ll add the option to change the aspect ratio because you’re stuck with 1:1 for now. An outpaint and upscaling option would be great too.
For now, this seems like one of the best AI image generators and best of all, it’s completely free.