OpenAI announced that DALL-E 3 is now available for ChatGPT Plus users. As revealed a few weeks ago, artificial intelligence to generate images from text was natively integrated into the popular chatbot. Users with a paid or business subscription will be able to access the new version starting today.
In an entry published on the blog of the company, OpenAI stated that have developed sufficient security measures to open its new model to more users. DALL-E 3 allows you to create images from natural language instructions (prompt) within ChatGPT. Compared to the previous version, intelligence has advanced considerably, especially in the interpretation of the parts of the human body.
According to OpenAI, DALL-E 3 can respond to extensive prompts and reliably renders intricate details, including text, hands and face. This was made possible by perfecting its subtitling technology, which is responsible for translating an input image into a textual description. Later, the company used it to train DALL-E 3, in this way, the model pays more attention to the instructions provided by the user.
An important detail is that DALL-E 3 now allows you to change the aspect ratio in your images. This was one of the constant criticisms from users, since the previous version was limited to creating 1:1 (square) content. It will now be possible to indicate the horizontal or vertical proportion in the prompt by ChatGPT.
To celebrate the launch of DALL-E 3, OpenAI showed examples of the images its artificial intelligence can generate. In one of them, the user asks ChatGPT to produce photorealistic images of clouds to be used for a school assignment. Another suggests a web design interface and logos.
DALL-E 3 will add more safety barriers
Generative AIs such as DALL-E, Midjourney or Stable Diffusion have been in the eye of the storm since their first versions. In addition to the copyright lawsuits that some of them face, there is also concern about the content they produce. Although many include safety railings, some manage to create offensive images.
With the arrival of DALL-E 3 on Bing, a considerable number of users found a way to bring back 9/11. A report of 404 includes illustrations of SpongeBob SquarePants and other characters piloting a plane over New York with a view to crashing into the Twin Towers. This forced Microsoft to adjust the guidelines to avoid this content within its search engine.
For the launch of DALL-E 3 on ChatGPT Plus, OpenAI revealed that They implemented a multi-level security system to limit the creation of potentially harmful images. That means We will not be able to generate violent, sexual or content that includes public figures from his name. The latter has been reinforced to prevent the proliferation of fake news, such as the arrest of Donald Trump or Pope Francis wearing Balenciaga.
The new version of artificial intelligence nor can it generate content in the style of living artists, a feature that aims to reduce criticism and legal problems. OpenAI has been criticized by the art community for training its model with copyrighted images without asking the original authors for consent.