[DigitalToday reporter Yoonseo Lee] OpenAI’s ChatGPT Image 2.0 generated more natural results than Google’s Nano Banana2 in a photo editing performance comparison, according to an assessment.

On April 28, IT outlet TechRadar reported that after comparing the two tools using prompts with high real-world utility such as portraits, product photos and changing seasons, the winner was ChatGPT Image 2.0.

The comparison focused on how naturally each tool could transform real photos rather than producing abstract or fantastical scenes. It examined which better maintained realism in edits that went beyond simple color correction, including changing backgrounds, adjusting lighting, and shifting mood and even seasons.

The method was simple. The same image was fed into ChatGPT and Gemini, and the same prompt was provided.

In the portrait test, ChatGPT Image 2.0 came out ahead. A request to change the background of a portrait to a “park with sunset light” felt somewhat unrealistic, but it was assessed to have relatively natural alignment in light direction and shadows, as well as how the chair and floor appeared grounded. Nano Banana2, by contrast, reflected the “sunset light” mood more strongly with deeper colors. But facial lighting did not match the strong sunlight in the background, and the skin rendering was overly smooth.

ChatGPT Image 2.0 also led in a test using the same portrait with a request to “make it look more cinematic and dramatic.” The ChatGPT output gave the impression of having been shot in an actual studio. Nano Banana2 interpreted the prompt by changing the composition and background more dramatically, moving toward a darker interior with bookshelves and wood panels. It was visually striking, but assessments said it felt like an “uncanny valley.”

Differences also emerged in product photos. When asked, based on an AirPods photo, to “stage it so it blends with a real background,” both models chose a desk background with coffee and a notebook. But the overall mood and style of the final output differed. ChatGPT Image 2.0 created the feel of a real workspace with sunlight coming in, and reflections on the AirPods surface connected naturally with surrounding objects. Nano Banana2, by contrast, was closer to a staged promotional shot. An earphone case that did not look like an AirPods case also appeared in the background, and it still looked awkward as an image meant to promote a specific product.

In the season-change test, there was little difference between the two models. When asked to turn a photo of green grass into an autumn mood, ChatGPT Image 2.0 left some green tones while creating a scene with fallen leaves scattered naturally. Nano Banana2 produced a more uniform and orderly result overall, but the colors of some trees and grass were still closer to spring. As a result, the category was rated a draw because it was hard to determine a clear winner.

Overall, Nano Banana2 showed strength in speed, but ChatGPT Image 2.0 received higher marks for naturalness. At first glance, the gap between the two tools did not appear large, but the closer the images were examined, the more differences emerged in the consistency of light, texture and depiction of people, the report said.

ChatGPT Image 2.0’s realism stood out in a blind test of non-experts. Participants were told that only one of several images was digitally edited, but they judged the ChatGPT Image 2.0 output to be a real photo. The test author assessed that ChatGPT Image 2.0 reproduced the interaction of light and texture, and how faces appear under different conditions, in a way closer to reality.

The comparison shows that evaluation standards in the competition among AI image generation tools are shifting from simple flashiness to realism in real photo editing. In areas with high commercial potential such as portraits and product photos, not only fast generation speed but also the naturalness of light, texture and facial depiction is emerging as key competitiveness.