Dall-E 3 rumoured to be on the horizon, and it can generate real text

In the ever-evolving world of artificial intelligence (AI), image generation is one area that continues to astound us with its rapid advancements. Recently, a video by YouTuber MattVidPro AI showcased a sneak peek into the future of AI image generation with what could be DALL-E 3, OpenAI’s latest iteration of their groundbreaking DALL-E model.

The video begins with Matt expressing his excitement about the cutting-edge capabilities of this new AI image generation model. He confidently claims viewers have never seen anything as impressive as what DALL-E 3 offers. He says it outperforms other prominent models like Midjourney, Stable Diffusion, and even the earlier version, DALL-E 2.

While DALL-E 2 was undoubtedly revolutionary when it was released in April of the previous year, the subsequent versions, such as Midjourney and Stable Diffusion, surpassed its image generation capabilities.

However, this did not deter OpenAI from further improving on DALL-E 2. The video’s exclusive content, obtained from a user on Matt’s Discord server, hinted at what DALL-E 3 might bring to the table.

One of the most striking features of DALL-E 3 is its ability to generate text coherently and accurately within images, which has historically been a challenge for AI models. The video showcases several examples of near-perfectly spelt and well-rendered text, from logos and product names to even complete sentences.

Despite its impressive advancements, the version of DALL-E 3 presented in the video is still a work in progress. OpenAI has removed safety features, enabling the model to generate violent and potentially upsetting imagery. While this unrestricted access has resulted in astonishingly realistic and detailed images, it also raises concerns about potential misuse or inappropriate content generation.

The video emphasized that OpenAI will likely fine-tune and add safety measures before any public release, ensuring that the model adheres to ethical guidelines and avoids generating harmful or offensive content.

We still don’t know when the update will be released, although MattVidPro AI speculated that it might happen before the end of the year.