OpenAI launches a new image generation model, challenging Google's picture-p

Among the latest developments in the tech world, OpenAI just announced that they have integrated the most advanced image generator to date in the latest GPT-4o model. OpenAI CEO Sam Altman excitedly shared his shock when he first saw the image generated by the model on social media platform X, thinking it was incredible and looking forward to users to get the most out of their creativity.

Highlights of the new features include:

- Ability to accurately render text content and provide high-quality image effects.

- Supports a variety of input and output methods, covering various forms such as text, images and audio.

- Understand complex instructions and combine context to create a first-person perspective image with a sense of reality.

Unlike the previous image generation model DALL・E, GPT-4o adopts an autoregressive model, natively embedded in ChatGPT. This means that it can handle complex instructions of up to 10 to 20 different objects, while competitors usually only handle 5 to 8, showing stronger capabilities.

Users simply need to describe their needs concisely, such as specifying aspect ratios, colors, or transparent backgrounds, and the model can quickly generate images. While rendering more complex details may take a moment, the final result is worth it.

At a press conference, the presenter presented several specific cases. For example, he transformed a group photo into an anime-style image. The model not only successfully retained the characters' characteristics, but also perfectly integrated the anime visual effects. In addition, the presenter asked to generate a page of humorous comics about the theory of relativity, and the resulting comics are not only complete in structure, but also vivid and interesting.

OpenAI also attaches great importance to the security of this feature. All generated images are marked with C2PA metadata, ensuring that the source of the content is traceable and effectively preventing the generation of inappropriate requests.

Of course, OpenAI's image generation tool is not without its shortcomings, such as lack of cropping, context understanding, and non-Latin text rendering. However, OpenAI said they will continue to optimize these issues in the future.

At the same time, Google also released its powerful AI model Gemini2.5Pro Experimental at the same time, showing a significant improvement in reasoning and programming capabilities. This series of dynamics shows that competition in the AI ​​field is becoming increasingly fierce, and major technology giants are constantly launching more advanced technologies, striving to occupy a leading position in this "AI battle".