ChatGPT Unveils New Image Generator: How Good Is It?

Could this be a reaction to the “code red” initiated by Sam Altman earlier this month? The OpenAI CEO shifted the company’s focus to better compete with Google and, to a lesser extent, Anthropic. Altman was particularly concerned about the superior performance of the Gemini 3 and Nano Banana Pro image generation models, which had recently topped specialized benchmarks.

Following the release of GPT 5.1 and 5.2, OpenAI has now turned its attention to image generation. In a blog post dated December 16, 2025, OpenAI introduced its new image generator within ChatGPT, named GPT Image 1.5. Here’s what we know!

GPT Image 1.5: OpenAI’s Latest Advance in Image Generation

Sam Altman can breathe a sigh of relief. The new model from OpenAI has quickly climbed to the top of the Text-to-Image Arena and the image generator section of the LMArena. According to the press release, it can “make precise edits while preserving details and generate images up to 4 times faster”.

Now, when you request modifications to an uploaded image, the model adheres to your intentions more reliably, capturing even the smallest details, and only altering what you specify, OpenAI promises.

Key enhancements include:

• Retouching: the model excels in adding, subtracting, combining, and repositioning elements to achieve the desired modifications.
• Transformations: GPT Image 1.5 transforms visuals by modifying and adding elements while retaining the important details of the original.
• Instruction Following: instructions are now followed more reliably, allowing for more complex compositions where the relationships between elements remain consistent.
• Text Rendering: the model now handles denser and smaller text content, including Markdown rendering and detailed infographics.
• Quality Improvements: the overall image rendering has been enhanced, particularly in the generation of multiple faces, such as in a crowd scene.

A New Feature in ChatGPT

The arrival of this model introduces a new section in ChatGPT, simply titled “Images”, located in the sidebar just below chat search. Here, you can access your images, along with suggestions for styles and formats. At the top, there is a prompt bar.

We tested the following prompt: “A famous actor responding to a crowd of journalists on the red carpet at the Cannes Film Festival”. The generation time is shorter than the previous generator, though still somewhat lengthy. The generated image is of higher quality, appearing more natural and without the yellowish filter seen in earlier versions. However, some abnormalities persist, particularly in the microphones held by journalists. Interestingly, ChatGPT seems to have created a look-alike of Tom Cruise.

We then requested a very specific modification: “Change the microphones to represent the following media outlets: France 2, Mediapart, LCI, and M6”. Previously, ChatGPT tended to generate an entirely different image. In our test, it indeed only changed the microphones, but the result was rough and inconsistent.

GPT Image 1.5 is now available to all users of ChatGPT, as well as in the API. OpenAI notes that it is “compatible with all models”.