OpenAI has launched its most advanced image generation technology to date, integrating the capability directly into GPT-4o, its natively multimodal model. The new feature is now rolling out to Plus, Pro, Team, and Free users in ChatGPT, with Enterprise and Edu access coming soon. Developers will also gain access via the API in the coming weeks.
OpenAI stated, “At OpenAI, we have long believed image generation should be a primary capability of our language models. That’s why we’ve built our most advanced image generator yet into GPT-4o. The result—image generation that is not only beautiful, but useful.”
Multimodal, Context-Aware Image Creation
The image generation tool in GPT-4o is designed to produce photorealistic and highly detailed outputs with strong adherence to user prompts. Built on a training dataset comprising both images and text, the model can generate visuals that communicate information clearly, such as diagrams, infographics, or posters, while also supporting more creative and artistic outputs.
GPT-4o is capable of generating complex imagery with up to 10–20 distinct objects, accurately binding objects to their traits and relationships. It supports in-context learning, allowing it to refine images across multiple turns in a conversation. For example, a user designing a video game character can iterate on their design while maintaining visual coherence throughout the process.
Precision and Practicality in Visual Communication
GPT-4o image generation excels at rendering text in images, enabling users to generate visual outputs that combine language and design with high precision. According to OpenAI, “From the first cave paintings to modern infographics, humans have used visual imagery to communicate, persuade, and analyze—not just to decorate.”
In addition to its ability to render symbols and structured data, GPT-4o can incorporate uploaded images into its generation process, using them for visual inspiration or transformation. This allows users to build upon existing content or maintain stylistic consistency across projects.
Limitations and Safety Protocols
OpenAI acknowledges that GPT-4o image generation is not without limitations. These include occasional cropping issues, hallucinated content in low-context prompts, challenges with precise edits, and difficulty rendering dense information or multilingual text. The company is actively working to improve these areas.
Safety remains a critical focus. OpenAI embeds C2PA metadata into generated images for provenance and uses internal tools to verify content origin. Requests that violate content policies, including those involving real people, nudity, or violence, are blocked by default. A reasoning LLM trained on safety specifications assists in moderating both input and output against policies.
“As with any launch, safety is never finished and is rather an ongoing area of investment,” the company noted.
User Access and Developer Integration
GPT-4o’s image generation will be the default for ChatGPT users starting today, replacing previous options. For those who prefer DALL·E, it remains accessible via a dedicated GPT.
Users can describe image specifications using natural language, including aspect ratios, hex color codes, and background transparency. Because the model produces more detailed outputs, images may take up to one minute to render.
Image: OpenAI