OpenAI announced that they will be releasing an updated version of DALL-E, named DALL-E 3, in October 2023.

Widely popularized for AI products such as ChatGPT, a Large Language Model (LLM) used by hundreds of millions of people overwide, and Whisper, a speech-to-text transcribing model capable of real-time translation, OpenAI has been a pioneer in the AI industry for the past few years.

Based on these prior AI technologies, the company has been developing DALL-E, an image generation AI, since early 2018.

DALL-E is capable of creating images based only on text input.

According to OpenAI, it can create a high quality image of “A modern architectural building with large glass windows, situated on a cliff overlooking a serene ocean at sunset,”,such as the image illustrated above, within seconds.

The upcoming DALL-3 would be the newest of the DALL-E series.

DALL-E 1, which was released in January 2021, was based on GPT-3, a LLM that had over 12 billion parameters at the time.

Not only that, but it was also trained on over four hundred texts and images from the internet.

After its release in 2021, DALL-E 1 was the most sophisticated image model to be built ever.

DALL-E 2, released a year after, served the same functions as DALL-E 1.

However, it was capable of creating images that had up to four times the resolution of DALL-E 1—a major improvement considering that the training difficulty for these models increased exponentially.

DALL-E 3, released almost a year and a half after DALL-E 2, takes a step further.

OpenAI states that DALL-E 3 is “built natively on ChatGPT,” and thus users would be able to use the model directly on the ChatGPT website.

Although the company has yet to release detailed statistics on how much of an upgrade DALL-E 3 is compared to the previous DALL-E 2, they claim that DALL-E 3 “delivers significant improvements over DALL-E 2.”

Furthermore, DALL-E 3 attempts to address some of the criticism DALL-E 2 received from the art industry, especially when it comes to copyrights and privacy.

Both DALL-E 1 and 2 were trained by “crawling”, or searching the internet for photos, and also enabled users to generate images with specific artistic styles.

With DALL-E 3, however, artists are able to opt out their artworks from training, and also configure DALL-E 3 so that it cannot replicate a certain artistic style.

OpenAI also states that they are aware of the fact that there is no way to differentiate an AI-generated image from a human-generated one, which potentially could lead to crimes such as identity fraud.

According to the OpenAI website, in an effort to address this issue, they are internally testing a system that can “help people identify when an image was created with AI” to reduce the confusion between the two.

DALL-E 3 will undoubtedly be a major advancement over DALL-E 2 with the ChatGPT integration and increased privacy and copyright measures.

Although there are criticisms regarding whether or not these measures are sufficient enough in fostering a better relation between the AI-image and art communities, it is certain that DALL-E 3 will change how generative art advances in the future.

Hoonsung Lee

Grade 11

Cornerstone Collegiate Academy Seoul