ChatGPT's stunning new image generator is now free for everyone

OpenAI has been steadily enhancing its ChatGPT platform, introducing features like an AI voice assistant, file and image understanding, advanced research tools, and AI agents. Yet, there was one notable gap in their offerings: a truly effective image generator. That changed last week when OpenAI unveiled its 4o image generation model, a significant step up from their earlier DALL-E models. Though it's a bit slower, 4o excels at handling complex prompts, including realistic images and precise text.

During a live stream demo, OpenAI's CEO Sam Altman, along with researchers Gabriel Goh and Prafulla Dhariwal, showcased 4o's capabilities. They prompted the model to create a photo from a specific point of view, featuring a flyer with detailed text. After a brief loading period, the result was impressive—the cinematic direction was spot-on, and all the text was accurately rendered.

Beyond its core functionality, 4o brings new features that weren't available in OpenAI's previous image generators. It includes image referencing, allowing users to transform an existing image into another style, like anime or a selfie, or use it as inspiration for a new creation.

this was a real labor of love from @gabeeegoooh. congrats gabe; excellent work!
here is what we generated during the livestream: pic.twitter.com/fmHWp4d9AF

— Sam Altman (@sama) March 25, 2025

Designed to integrate seamlessly into creative workflows, 4o can generate images on transparent backgrounds, use specific colors via HEX codes, and even incorporate the chatbot's advanced conversational abilities into the image creation process. For instance, when asked to add humor to a photo during the demo, 4o included text that reflected that request.

The image generator is now part of ChatGPT, enabling users to refine images through ongoing conversations. This feature simplifies tweaking images and allows the model to leverage the context of prior generations for new creations. With GPT-4o's access to the web, this context further enriches the image generation process.

OpenAI claims that GPT-4o's image generation model has strong adherence to instructions, capable of handling 10-20 different objects simultaneously, allowing for complex scenes to be created in one prompt.

Looser Safeguards

A notable change with the new image generator is its ability to create more risqué content, similar to what Elon Musk's Grok model is known for. During the live stream, Altman mentioned that users would be able to generate offensive content "within reason" using GPT-4o. Following the event, Altman elaborated on X:

"What we'd like to aim for is that the tool doesn't create offensive stuff unless you want it to, in which case within reason it does. As we talk about in our model spec, we think putting this intellectual freedom and control in the hands of users is the right thing to do, but we will observe how it goes and listen to society."

The model's announcement blog post clarified that it will block requests violating content policies, such as child sexual abuse materials and sexual deepfakes. Additionally, there are robust safeguards when real people are involved, especially concerning nudity and graphic violence. Users can refer to the System Card for detailed safety information on the 4o image generation model.

How to Access

The updated image generation features are now rolling out in ChatGPT and Sora, accessible to all users, including those on free accounts. However, to fully utilize GPT-4o, you need to use the "/create image" shortcut. Simply typing "Create an image of XYZ" will default to the older DALL-E model, resulting in lower-quality images.

OpenAI hasn't specified limits on image generation, but after creating three images on my free account, I hit the daily limit. For more extensive access, subscribing to ChatGPT Plus at $20 per month is the best option for individual users, offering additional benefits like access to OpenAI's Sora video generator. At the time of writing, I was able to use the image generator with my Plus account. Enterprise and Education users will gain access soon, with developers getting API access in the coming weeks.

When DALL-E first debuted, it was a standalone marvel. Now integrated into ChatGPT, it had lagged behind competitors like Midjourney, Google, and Adobe. This update with 4o helps level the playing field, making it more competitive. However, if you still want to use the original DALL-E, it's available through a dedicated DALL-E GPT.