OpenAI is reportedly running a series of beta tests to test the forthcoming version of their DALL-E AI text-to-image generator. A fictitious source in Discord has released information about the tests. The leaker claims they have been a part of beta testing of OpenAI’s forthcoming AI image model, DALL-E 3. The leaker also distributed images that were generated in the period of testing.
Leaked data suggests the beta version released in May of DALL-E 3 could create images with multiple aspect ratios. Also, it showcased the model’s remarkable text generation abilities, which addresses a frequent problem that other models have to face.
The photos generated in the experiment included a variety of examples of the text incorporated in a brick wall, neon sign, billboard, cake decoration, and an inscription of the name on the mountain. The model also showed the capability to produce pictures of people. This included one of a woman consuming spaghetti during a dinner party in a completely different way.
The leaker resurfaced on the Discord channel in mid-July and provided further details and images. The leaker claimed they were part of a beta closed test involving around 400 people. These changes indicate substantial improvements in DALL-E 3 since May. The leaker posted images from the same request, demonstrating how the model’s capabilities changed.
Compared to other models, such as Midjourney, DALL-E 3 showed better performance and produced higher-quality and more detailed pictures based on the request. The leaker has also stated that the beta testing model for DALL-E 3 has the potential to make explicit, copyrighted material, which includes images of nudity and violence.
DALL-E 2, the previous version of OpenAI’s AI image generator released in April 2022 but was tightly controlled and available only via waiting lists. It was released for public use on September 20, 2022, in response to fears about the ethics of safety and security.
The information leaked provides insight into the latest developments in the DALL-E AI model and its possible applications throughout various sectors.
