Comparing chatbots has become common practice, with users posing identical questions to different systems to assess accuracy and intelligence. That curiosity has now extended to AI image generation, a field long criticised for producing awkward visuals often dismissed as “AI slop.”
For years, artificial images have flooded the internet with telltale flaws. Blurred backgrounds, overly smooth faces, and anatomical errors have given AI-generated visuals a poor reputation. Stock photography, however, is now seeing a noticeable shift.
A major change in this space followed OpenAI’s recent upgrade, which enhanced image generation within ChatGPT. The update has positioned ChatGPT as a stronger competitor to Google’s Gemini Nano Banana Pro, particularly for stock-style visuals.
The most striking improvement is realism. Heavy blur, plastic-like textures and exaggerated lighting are now far less common in ChatGPT’s output. As a result, images feel more practical for articles, advertising and social media use.
GPT Image 1.5 didn’t kill Nano Banana Pro. Seedream 4.5 did.
I tested three of the strongest AI image generators for realistic human portraits:
– Seedream 4.5
– GPT Image 1.5
– Nano Banana Pro
All three can generate good-looking images. Only one can preserve identity.… pic.twitter.com/nLAMCiCRxF
— Beginnersblog (@beginnersblog1) December 22, 2025To test both systems, identical prompts were used across everyday scenarios. When asked to generate an image of someone listening to a podcast in an office, Gemini Nano Banana Pro produced a more convincing result. Subtle imperfections, such as slightly untidy hair and a softer background, made the image appear more authentic. ChatGPT’s version was clean and quick to generate, but it still felt overly polished and somewhat artificial.
A similar pattern emerged with a prompt featuring a person reading a book for promotional use. Gemini Nano Banana Pro again leaned closer to realism, while ChatGPT struggled with staged lighting and poses that revealed a computer-generated look.
The new image generation model in ChatGPT adheres more reliably to your intent, changing only what you ask for while keeping elements like lighting, composition, and people’s appearance consistent across inputs, outputs, and subsequent edits.
It also excels at different types of…
— OpenAI (@OpenAI) December 16, 2025ChatGPT, however, performed better when tasked with creating a generic stock image for editorial use. Its result, while not entirely natural, was usable. In contrast, Gemini Nano Banana Pro failed to meet the brief, with an overly conceptual design and an abstract.
In a final test focused on leadership challenges, both tools delivered familiar office scenes that lacked creativity. Even so, Gemini Nano Banana Pro once again edged ahead on realism.
AI image generation continues to evolve rapidly. While neither system is flawless, Gemini Nano Banana Pro currently holds a slight advantage in lifelike detail. ChatGPT, however, is rapidly closing the gap. At this rate, the distinction between real photography and AI-generated stock images may soon become almost impossible to detect.
