OpenAI has added image generation to ChatGPT with improved prompt understanding

OpenAI has introduced the ability to generate images directly within ChatGPT. The feature is based on the omnimodal GPT-4o model, capable of processing text, images, audio, and video.

This allows for more accurate interpretation of user prompts and the creation of more complex scenes.

Unlike DALL-E, the new system uses an autoregressive approach in which the image is generated gradually — line by line and left to right, similar to writing text.

Although generation takes more time, OpenAI believes the quality justifies the delay. The free usage limit remains the same as DALL-E's but may vary depending on server load.

All images are embedded with C2PA metadata and do not contain visible watermarks. Abuse protection has also been strengthened: fake and harmful content is blocked.

It was previously reported that OpenAI is preparing to launch a new AI model called "Strawberry" within the next two weeks, although the release was initially planned for the fall.

The developers added a 10–20-second pause before responses to improve accuracy and handle complex analysis better.

The model supports only text but can remember conversations for personalized replies — though this feature remains unstable. "Strawberry" will be offered as a separate option in ChatGPT with a message limit.