OpenAI gears up for DALL-E 3 launch: A superior Image Generator surpassing competitors and evading censorship rules

OpenAI is reportedly developing an upgrade to its DALL-E 3 image generator language model, called DALL-E 3. This version, currently accessible to around 400 users, surpasses the previous model from OpenAI and recognized competitors like Midjourney or Stable Diffusion, according to those who have had access to the preview.

YouTuber MattVidPro AI demonstrated the capabilities of DALL-E 3 in a recent video, showing that it can generate images that resemble photos or complex paintings without following censorship rules. This includes the generation of violent and copyrighted content (This is likely to be limited in the final version when it is released to the public). For example, he created a painting of a zombie Jesus Christ eating a sandwich from Subway, in a style quite similar to Francisco Goya's painting "Saturn Devouring One of His Children".

The YouTuber highlighted DALL-E 3's improved understanding of prompts and text generation, particularly in generating images of hands. Compared to Midjourney's latest model, DALL-E 3 adheres better to prompt rules and provides higher image quality. This development of course challenges companies like Midjourney, Microsoft, Google, and Adobe to upgrade their own versions of image generators language models in order to keep up with OpenAI's progress.