Recently, the latest text-to-image model from OpenAI, GPT Image2, has shown strong performance in authoritative evaluations. According to the latest data released by SuperCLUE, the model has officially surpassed Google's Nano Banana2 and successfully claimed the top position in global text-to-image model evaluations. It is reported that since its launch on April 21, the model has significantly improved in image quality, comprehension, and detail restoration, setting a new industry standard.

In this evaluation, GPT Image2 demonstrated comprehensive performance across multiple core dimensions. Especially in the field of Chinese character generation, which has long been a challenge for overseas models, the model achieved a high score of 93.07, with text accuracy receiving a perfect rating. It not only can accurately identify and generate complex Chinese characters, but also achieve an integration of text with different material textures such as acrylic and blue and white porcelain, effectively solving technical issues such as the "floating" feeling of text and character corruption.

1.png

Aside from breakthroughs in text processing capabilities, the model also demonstrated a high level of instruction following in replicating complex scenarios. From a traditional bread shop full of life to a dynamic intangible cultural heritage iron flower display, GPT Image2 can accurately capture visual details. In addition, for long prompts and logical reasoning requirements, the model can accurately produce high-difficulty content such as scientific principle diagrams and professional posters, demonstrating excellent text-image consistency.

Although the evaluation report also pointed out that GPT Image2 still has some room for improvement in understanding spatial relationships and deep knowledge reasoning, its advantages in realistic reproduction and creative reasoning are sufficient to make it stand out in the competition with competitors such as Google and Baidu.

Industry experts believe that the release of GPT Image2 not only marks OpenAI's continued leadership in the field of visual generation, but also indicates that text-to-image technology is moving from simple imagery generation to a professional stage characterized by high precision and logical emphasis. With continuous model optimization, the boundaries of AI visual creation will be further expanded.