Now, image generation feels like having a conversation.
On March 25, 2025, OpenAI unveiled a more natural and sophisticated visual creative tool by integrating GPT-4o’s image generation feature directly into ChatGPT. This update replaces the existing DALL·E-based system and allows AI to understand user sentences and generate images according to the context of the conversation.
ChatGPT users can now create images directly in the chat window without any additional tools. It is available across all Plus, Pro, Team, and Free subscription plans, with free users limited to generating up to three images per day. Paid subscribers can create images without a limit, and enterprise and educational users will soon have access to this feature as well.
OpenAI plans to release an API for developers in the coming weeks, allowing external apps and services to leverage GPT-4o’s image generation capabilities.
It can accurately depict complex scenes.
GPT-4o has advanced processing capabilities that can recognize and arrange up to 20 objects simultaneously. For example, it can naturally construct a complex scene like “a dog watching a movie while eating popcorn.” Difficult-to-render elements such as hands, faces, and text are rendered in detail, and the image generation speed ranges from an average of 30 seconds to a maximum of one minute.
The reporter’s test also showed impressive image generation capabilities of GPT-4o. When the sentence “Create an image of a dog watching a movie while eating popcorn” was entered into ChatGPT, the scene was rendered exactly as requested within seconds. The image of the dog sitting on a sofa, holding popcorn and looking at the screen, was intricately composed, including challenging elements like the hands and face, which were rendered naturally.
Building on this image by adding speech bubbles and background, a four-panel comic was produced without disrupting the image quality or style. The ability to create comic images solely through text input offers a fresh shock to traditional content creation methods.
Users can precisely specify colors (hex code), background (including transparency), ratios, and freely dictate desired styles and detailed elements in natural language before generating an image. The generated image maintains the context of the conversation, allowing for repeated generation or modification requests in the same style.
OpenAI emphasizes that this method evolves beyond simple AI functionality into a ‘generative interface’ that combines creation and communication.
DALL·E transitions to a supplementary role.
With this GPT-4o integration, DALL·E steps down from being the primary image generation tool. However, OpenAI retains DALL·E as a separate GPT-specific option for its strengths in artistic and experimental images, allowing users to switch freely between the two models as needed. Users can choose GPT-4o for practical and commercial work and DALL·E for emotional and artistic tasks.
This provides users with a wide range of options to balance functionality and creativity, enabling the use of generative AI tools tailored to various professions and purposes.
A generative ecosystem connected to Sora.
OpenAI is also integrating the image generation feature with its video generation model, ‘Sora.’ Sora is an AI video generation engine that can take text, images, and videos as input to create 20-second videos in 1080p resolution.
Users can input scene settings in a storyboard format, edit existing videos, or seamlessly connect two videos. Currently, Sora is offered to Plus and Pro plan subscribers, with Pro users able to download videos in high quality without watermarks.
Sora is being utilized in content production fields such as advertising, short films, and animations, and some Hollywood producers are reported to be testing it.
OpenAI is building a ‘multimodal environment’ in which generative AI functions as a single creative tool across text, image, and video domains through the integration of GPT-4o and Sora.
This update demonstrates that image generation technology is no longer just a technical novelty but is becoming a central tool for actual creation and communication. OpenAI states, “The goal is to ensure wider creative freedom for users while appropriately controlling according to social standards.” All images generated include C2PA metadata for attribution, enhancing traceability and transparency.