“Tubeli-safe” like ChatGPT: OpenAI presents the new AI image generator DALL-E 3

The American company OpenAI, developer and supplier of ChatGPT, releases a powerful new version of its AI image generator: “DALL-E 3”.

The technology portal futurezone.at says:

What does DALL-E 3 offer?

The most important innovation is the integration of ChatGPT. It should now be enough to enter a few simple sentences so that the image generator can create convincing “works”. And the context of the user input is taken into account.

OpenAI promises:

This means that in the future no complicated search for the right formulation, also known as prompt engineering, will be necessary to obtain convincing results.

Below are some examples, which also come from OpenAI itself and were published by X. What’s particularly interesting is the relatively short ‘prompt’, i.e. the text input used to generate the images by the AI.

When and for whom is DALL-E 3 available?

The new version of DALL-E-3 will first be integrated into the paid ChatGPT Plus and ChatGPT Enterprise plans and should be available “in October.” It is unclear when the new image generator can be used for free.

How safe is the new image generator?

According to its own information, OpenAI has integrated 3 security mechanisms into DALL-E that are intended to prevent the image generator from spewing out “hateful messages.” To achieve this, we have collaborated with external experts. The AI ​​has learned to ignore certain words that contain violence. The names of some celebrities also ended up on this blacklist. Accordingly, it would no longer be possible to generate images of famous personalities, as was the case with the Pope.

Then there is the issue of copyright infringement by generative AI, which OpenAI wants to prevent with additional safeguards. More information about futurezone.at:

The Verge reminds us that lawsuits are already underway in the US. DALL-E competitors Stability AI and Midjourney, as well as art website DeviantArt, were sued for allegedly using copyrighted works to train their AI image generators.

Sources

(dsc)