It is reported that OpenAI is testing the DALL-E 3 model, which is expected to promote AI image generation to a new stage.

CTOnews.com August 2 news, OpenAI in April last year launched the second generation of DALL-E "Wensheng diagram" model, the model with excellent strength attracted wide attention in the industry, and therefore once out of the circle.

According to foreign media Decoder, OpenAI is preparing the next generation DALL-E AI model (DALL-E 3). At present, the company is conducting a series of Alpha tests, and some users have been exposed to the AI model early.

According to foreign media, an anonymous user on Discord recently shared his experience in testing DALL-E 3.

The anonymous user said that the May test version of DALL-E 3 had been able to generate images with multiple aspect ratios, support longer prompt statements and generate "normal text".

▲ diagram source Decoder

The July test version is even more powerful, but the anonymous user said that the July test version may randomly generate a series of NSFW scenes, as well as copyrighted trademark patterns, and OpenAI hopes to ban the generation of such "inappropriate content" at the model level, so DALL-E 3 still needs some time to make relevant training adjustments.

Foreign media said that DALL-E 3 has made a breakthrough in image quality and diversity, which is expected to promote AI image generation to a new stage.

CTOnews.com reported earlier that Stable Diffusion XL has launched the official version 1.0, and Stability AI claims that Stable Diffusion XL 1.0 is the company's "most advanced" AI tool so far, and the actual effect is indeed remarkable, so OpenAI's DALL-E model will also have a "head-on challenge" with it.