You may remember when, a few months ago, half the internet started creating Studio Ghibli-style images with ChatGPT to upload them to social networks. The “magic” behind that fever was OpenAI’s new image generation model. But everything moves so fast that the conversation lasted just long enough. Some time later, our attention was already elsewhere: on how difficult it was to distinguish some images created with Nano Banana Pro.
In a very short time, half the world began to talk about the benefits of Google’s new generative model, and there were many who placed it ahead of that of OpenAI. But this is an open race, with technology giants fighting for AI leadership. And, as expected, the company led by Sam Altman has responded. This Tuesday it launched ChatGPT Images, which comes boasting several improvements for users.
Editing as a key element. One of the great historical challenges of image generation tools has been the specific editing of specific elements. ChatGPT Images directly aims to solve this limitation, allowing us to modify only what interests us, from a specific object to lighting, composition or even the appearance of people. This opens the door to combining elements or introducing very specific changes without having to remake the entire image, something that until now used to be a weak point in this type of model.
See yourself in an ad or “travel” to your favorite place. Another section where ChatGPT Images makes a leap is in creative transformations. Simply upload your own photo and accompany it with a simple prompt to obtain, in a matter of seconds, surprisingly convincing results. It is worth clarifying that this idea is not completely new. In fact, it is one of the most outstanding virtues of Nano Banana Pro, a model that our colleague Javier Lacort was able to test thoroughly and was already pointing in this direction.
Let’s look at some examples with ChatGPT Images.
Original image:
“Create an image of this man, but in Time Square, New York, with clothes, looks, surroundings, etc., that are believable for winter 2025”
“It places this person in full body in a Japanese city during a rainy night, with neon, reflections on the ground and cyberpunk aesthetics”
Precision as a flag and improvements in the text. OpenAI also places emphasis on improving accuracy. How many of us have had to ask for something specific and receive just the opposite, or find that the model has not understood the instruction correctly? Part of that problem, according to the company, should be left behind. If we provide detailed instructions, the system should be able to honor them more faithfully. In addition, the generation of text within images is reinforced, a key aspect for creating posters, promotional ads and other content where typography and the message are as important as the image itself.
Images | OpenAI
In Xataka | We believed that Microsoft had already put Copilot everywhere. LG shows us that we were very wrong
GIPHY App Key not set. Please check settings