In April, OpenAI’s DALL-E 2 system was unveiled, offering a text-to-image generator that has been trained on robust datasets featuring millions of images and associated captions. While the project is not publicly available just yet, Boris Dayma programmed a stripped down version of the AI art generator, known as the DALL-E Mini, that has apparently been discovered by Twitter this week — en masse.
To be clear, the images generated by DALL-E Mini do not have the same quality or detail that DALL-E 2 would be able to create. The latter is able to produce fairly high-resolution images (1024 pixels x 1024 pixels) at lower latency.
While DALL-E 2 is still being fine-tuned, the program won’t likely be widely available even when it releases — at least upon its initial launch. Either way, we’ll have the DALL-E mini to hold us over.