Chatgpt premieres a function to create images from scratch and is surprisingly good

Openai announces the launch of a function within chatgpt for the generation of images, aimed at Users Plus, Pro, Team and those who enjoy the Chatbot free of charge. The tool is part of the GPT -4o multimodal model and can be used within Sora.

“The generation of Images GPT -4O stands out for the precision in the text representation, the exact monitoring of indications and the use of the knowledge base and the context of the GPT -4O chat, which includes the transformation of loaded images or its use as a visual inspiration,” says the official statement of the company. The function allows to generate high quality content in response to conversations, Prompts (instructions) and uploaded files. Specifically, users can create images from scratch or modify existing photos according to the indications that provide to the Chatbot.

A picture is worth a thousand words

According to OpenAI, the highlight of the generation of GPT -4o images is the capacity of the model to combine specific symbols with visual content. From logos to infographics, the images can transmit a precise meaning when complementing symbols that refer to shared language and experience.

GPT -4O follow detailed instructions with detail. While other systems have difficulties with between 5 and 8 objects, The GPT -4O generator can manage up to 10-20 different objects. The greatest linking of objects with their characteristics and relationships allows better control.

OpenAI

It is clear that Openai’s intention is to improve the ability of its image generation model, offering its users a wide range of options. In the next few days, it will be possible to use the Chatgpt generative AI to create diagrams, infographics, promotional graphics for social media, logos, invitations, etc.

How to identify if an image was generated with AI

A brush is a tool that humans use to create artistic expressions. AI systems are a bit more sophisticated, they create almost “perfect” images, difficult to label as synthetic. Here are some tips to identify the real of fantasy.

Limitations and security policy

On the other hand, the company founded by Sam Altman refers that, to avoid security conflicts, all generated images have C2PA metadata, which will identify an image as coming from GPT -4O, providing transparency. “We have also developed an internal search tool that uses generations technical attributes to verify if the content comes from our model,” explains the statement.

In order for the function to be safer, Openai promised to limit the generation of images that “can violate content policies, such as child sexual abuse and Deepfakes sexual “. Functional, practical and safe, the new GPT -4o model tool will facilitate the generation of images.

Article originally published in Wired Italy. Adapted by Alondra Flores.