4o model: New image generator in ChatGPT

Creating images with ChatGPT will also be possible in future using input from conversations and uploaded files. Both generation and image processing have been improved, OpenAI announces. This applies to following instructions, i.e. how exactly prompts are implemented, context understanding and text rendering. The reason for this is the conversion of the image generator to the 4o model. Dall-E will be replaced accordingly. However, the separate image generator will continue to be available.

With the switch to the omnimodal model 4o, image generation will become native, which means that two different models will no longer be responsible for text and images. All people with Plus, Pro, Teams and Free accounts will have access. However, the introduction is happening gradually. Enterprise and Edu customers will have their turn later.

Better logos, writing and editing

OpenAI promises significant improvements to imagesin the blog post. This includes, for example, more accurate image generation of diagrams, infographics, logos and promo graphics for social media with hexadecimal codes. As text can be reproduced much better, even business cards can be designed, writes OpenAI. It is also possible to create images with a transparent background, which can then be incorporated into presentations, for example.

Images can be modified on the basis of a template. OpenAI is thinking here of interior design ideas based on a living room photo, for example. As an example of generating an image based on a conversation, OpenAI mentions the bird species in Central Park or the visualization of an era that is currently being discussed.

When introducing GPT-4o, OpenAI had already said that the model could understand and generate text, audio and images at the same time. This means that information no longer has to be passed from one model to another. For example, previously one model had to generate text and pass it on to another model in order to turn it into an audio file. This is a potential source of error.

Images generated with OpenAI's tools always receive a reference to this in the metadata. OpenAI uses the open standard C2PA for this, which is also used by camera manufacturers, among others, to verify the origin of a photo.

OpenAI CEO Sam Altman writes on X that the new image generator gives users more freedom when generating images. In other words, less content will be denied. "We think it's right to put this intellectual freedom and control in the hands of the users, but we will watch how it develops and listen to society." He also believes it is right to respect the "very broad boundaries that society will ultimately set for AI, and that it will become increasingly important the closer we get to AGI."

This brings OpenAI closer to Elon Musk's image generator, which is integrated into Grok. This also has hardly any guard rails when it comes to the creation of images.

(emw)