The GPT-4.5 feature of images generation appeared

On March 25, 2025, Openai announced the integration of the generation function of the images into the updated version of the Chatbot ChatGPT-GPT-4.5. The statement says that modern generative models can create “fantastic scenes”, but it is difficult for them to reproduce useful images.

It is known that the GPT-4O generates the photo, focusing on the given text parameters, following the instructions and using knowledge of both the context of the chat and from the loaded images.

Also, Openai called the advantages of the updated version of GPT-4.5, namely:

• Reproduction of the text. GPT-4O can combine characters with visual content;
• A multi -stage image improvement. Users can change and adjust images in several stages through a dialogue;
• Accurate following instructions. GPT-4O is able to process requests with a detailed description and place up to 10-20 objects in the image, while most models are limited to 5-8 elements;
• Training in context. The model can analyze loaded images and take into account their details when creating new pictures;
• World knowledge. GPT-4O integrates text and visual information, which makes its generation of images even more intelligent and effective;
• Photorealism and style. The model is trained in images of various styles, which allows you to create or convert images as convincingly as possible.

The report says that the GPT-4O is not perfect and has certain restrictions that Openai plans to eliminate in the following updates. In addition, each created image will contain data that will indicate its origin from the GPT-4O. Openai also developed an internal search system that allows you to check whether the image was generated by this model.

It is known that the GPT-4O will block unacceptable materials and images, in particular, containing violence, pornography or manipulative deep fakes.

In addition, developers will be able to use the generation of images in the GPT-4O through the API in the coming weeks. Users are also available for a flexible setting of images – you can change the ratio of sides, color and background transparency. The Dall-E model can be used in the future, the statement said.

Be in the know! Subscribe to Telegram.