OpenAI Launches Free Image Composition Model Powered by GPT-4o

OpenAI has officially introduced a new image generation feature in ChatGPT, powered by the GPT-4o model.

Unlike the earlier DALL·E 3, this new system has a much deeper understanding of context, handles complex instructions with precision, correctly links attributes to objects, and can generate legible text directly on images without distortion. The tool is called 4o Image Generation and is rolling out today to all ChatGPT users — including those on the free tier, with a limit of up to three images per day, similar to DALL·E.

The standout feature of the new system is its compositional generation approach — images are rendered gradually, left to right and top to bottom. This allows for more accurate detail placement, especially when working with text or multiple elements.

For instance, when asked to create a scene featuring a dozen distinct shapes and colors, GPT-4o handles it flawlessly — whereas other models often stumble by the fifth or sixth element. It's also improved at handling transparent backgrounds, making it especially useful for generating logos, stickers, and UI assets.

Another notable upgrade is the ability to edit existing images. Users can ask to add or remove elements from a photo using just a text prompt, and the system will apply the changes accordingly.

OpenAI has also emphasized its safety measures. All generated images are tagged with C2PA metadata to indicate their origin, and the system blocks attempts to generate deepfakes, violent content, or other sensitive material. Developers say they’ve put a particular focus on abuse prevention, especially in light of recent scandals involving deepfake misuse and misinformation.