Generative AI has moved from novelty to essential creative infrastructure in just a few years. In early 2023, the world marveled at two breakthrough abilities: AI could write comprehensive content, and some models could generate graphics and photorealistic images. By 2026, the market for AI image generators has matured into a competitive field with dedicated paid tiers, advanced editing tools, and increasingly sophisticated outputs. A new evaluation of six popular AI image generators set out to determine which tools are worth a monthly subscription. The results are unusually decisive.

The evaluation generated more than 50 images across six platforms. Each generator was subjected to 30 individual evaluations, producing an aggregate score. Nine tough prompts covered four main categories: revising existing images, generating original images, adding text to images, and leveraging pop culture. All images were created in a 1:1 square ratio for easy side-by-side comparison. The outcome: a no-contest, total knockout winner.

Key Takeaways

• Google's Nano Banana Pro, part of Gemini 3, earned a near-perfect score of 93%.
• ChatGPT image generation ranked second at 74%.
• Midjourney, Adobe Firefly, Leonardo AI, and Canva scored between 43% and 57%.
• Text rendering, photo restoration, and pop-culture prompts separated the leaders from the laggards.
• Paid tiers often differ mainly in usage limits, though quality varies dramatically between platforms.

The Clear Winner: Google Nano Banana Pro

Nano Banana Pro, available through Google AI Pro for $19.99 per month, is the standout champion. It aced every test with a near-perfect 93% score. No other tool came close. The evaluation called it a first in AI image generation testing: usually, several tools perform fairly well, but here one model dominated across nearly every category.

Photo recontextualization was one of the first tests. The evaluator fed the AI an existing image and asked it to place the subject in a completely different setting. The formal test involved taking a picture of a person out on a walk and asking the AI to put him in a US Navy admiral's uniform on the bridge of an aircraft carrier. Nano Banana Pro kept the face and glasses, changed the angle the subject was facing, placed him properly on the bridge, and even added binoculars to his hands. There were minor issues with Navy personnel and beards, but the overall result was highly convincing.

Next came old photo restoration. The evaluator found a childhood photo taken on the USS Ling that was too dark to see. Nano Banana Pro cleaned it up, brightened the image, and preserved the original features. The result was a clear, modern-looking restoration. A separate colorization test used an image of a truck. Nano Banana Pro mostly preserved the lettering, though it introduced a typo: BADICLOGICAL DEFEKSE instead of RADIOLOGICAL DEFENSE. That cost a few points. The truck was also colored like many emergency vehicles of the era, which may or may not have been historically accurate.

Creative generation showed the model's range. For a logo for Space Coast Studios, Nano Banana Pro picked up all the themes: a palm tree for Florida, a rocket, film imagery, and clean, readable text. For a fantasy librarian in a candlelit stone library, it produced a detailed scene with candles, stone walls, and appropriate atmosphere. Social media tests also performed well. Asked to create a senior adult holding a phone with short ad copy, the model generated a realistic image with properly placed text. It also created a student in a cozy environment with a MacBook Pro and a cup of coffee. The text was correct, fit cleanly, and had no errors. One amusing detail: Google's AI generated what was clearly an iPhone when asked for a flagship smartphone.

Pop-culture prompts were another strength. For a fictional fourth Back to the Future movie set in 1920s New York, Nano Banana Pro included the DeLorean and skateboard, evoked the original logo style, placed New York City 1925 on a marquee, and added the tagline The Roaring Twenties Got Heavy. It also dressed Marty McFly in period clothing. For a Nightmare Before Christmas style IT professional in a data center, the model delivered a strong Tim Burton-inspired image. Across all tests, Nano Banana Pro made only minor errors: one session confusion during the truck test and a watermark in the bottom corner of each image. The watermark can be removed with another AI, but it remains an annoyance.

Second Place: ChatGPT

ChatGPT image generation, part of ChatGPT Plus at $20 per month, ranked second with 74%. The tool is officially described as GPT-4o image generation in ChatGPT, and OpenAI later introduced a ChatGPT Images update. The new version improved in some areas but also lost ground in others. The previous version scored 81%, while the new version dropped to 74%.

Photo retouching is not a strength. In the admiral recontextualization test, the face looked like the original subject, but the uniform had unclear rank insignia that seemed to mix a Lieutenant and an O-8 Admiral. The childhood photo restoration was good: the original face was preserved, unlike an earlier version that put a random adult face on the child's body. The truck cleanup was decent, but the lettering changed from RADIOLOGICAL DEFENSE to RAD

Source:ZDNET News