ChatGPT’s New AI Image Creator 🎨

Mar 28, 2025

Ask episode

Chapters

Transcript

Episode notes

OpenAI just made the most significant leap in image generation I've seen over the past year. You can now type a simple prompt using ChatGPT 4o and create a remarkable photo illustration, infographic, cartoon, or just about any other visual.

What makes this special?

* Versatility. Create nearly any kind of visual you can imagine.

* Intelligence. The AI understands your intent based on an ongoing chat thread and its understanding of the world, rather than just focusing on prompt phrases. That means you don’t have to master technical lingo or explain common concepts.

* Continuity. You can create variations on any image and use consistent characters or styles for ongoing stories, presentations, or projects.

* Text. I’ve been amazed at the rendering of vast amounts of text inside images, as in the parking sign above. Other AI tools struggle with more than a few words.

It’s available for all ChatGPT users, whether you’re on a free or paid plan, on any platform. Read on for how to make the most of it, limitations, and alternatives.

7 ways to use ChatGPT's new image AI

Cartoons

I've always wanted to draw cartoons but never had the skill. Now I can quickly prototype visual sequences. While human cartoonists bring unique creativity that AI can't replicate, this tech allows anyone to experiment.

Infographics

What impressed me most as I beta tested this model in recent weeks was its extraordinary level of nuance, detail and text accuracy. I created explanatory infographics for AI learners and music appreciation students. If you've spent hours building infographics or relied on stock, this may be a turning point. Caveat: The model sometimes struggles to accurately render text in non-Latin languages.

Posters

Create event ads, announcements, social posts, or signage without having to rely on a template. Quickly test out visual ideas that might otherwise take hours to flesh out.

Slides

Generate compelling images for presentations. Create wide or tall slides with big words or numbers, stylish quotes, or clarifying flowcharts. You can now use ChatGPT for help with planning a deck and designing its slides. Determining the purpose, structure, style, approach, and delivery is still your human role.

Illustrations

While DALL-E 3 (ChatGPT's previous image tool) worked well for some illustrations, this new 4o image generation opens up a broader range of styles, including conceptual images (like this) for blog posts, newsletters, or videos.

Stories

If you write fiction or poetry, you can now generate consistent character images. I’m delighted to be able to experiment with illustration styles for fan fiction I’m working on with my daughters based on the "Not Quite Human" series about a robot disguised as a human teenager.

Designs

Create icons, logos, or micro-illustrations for your projects. You can ask for multiple versions of a design in different styles, then build on the one you prefer.

How to prompt ChatGPT 4o for great images

1. Iterate through conversation Unlike other image generators that require a new prompt each time, ChatGPT 4o now enables an ongoing revision dialogue. Ask it to change styles, adjust elements, or create multiple related images. Caveat: asking for a correction on one element sometimes results in unexpected changes to other parts of an image. And ChatGPT will refuse some requests on content policy grounds.

2. Upload reference images The multimodal nature of the model helps it understand and incorporate elements from images you share. I uploaded an image from a well-designed invitation and used it as inspiration for a private book group visual.

3. Prompt for prompts Use ChatGPT, Claude, or Gemini as a thought partner to suggest effective ideas or prompts based on your goals. This meta-approach helps you broaden your ideation.

4. Compare across services Even with this major advancement, it's worth testing your prompts in other services, like the ones noted below, to see how results differ and which model works best for a particular project.

5. Save winning prompts When you find a formula that works well for the kind of images you'll want to generate repeatedly, save it. A snippet manager like Raycast, Alfred, or TextBlaze makes saving and reusing these prompts easy. Once created, you can just type "\illo" — or whatever keyboard shortcut you choose — to paste in your favorite illustration prompt. This allows you to add custom details while keeping your base prompt intact.

Limitations

OpenAI has acknowledged several technical limitations of the new image generation model in their surprisingly candid launch post.

1. Cropping challenges When creating wide or tall infographics or slides, the AI sometimes misjudges dimensions, resulting in cut-off text or images. You may need to prompt again to fit all content properly.

2. Complex information hallucinations For complicated requests like showing all elements in the periodic table, ChatGPT may struggle to track more than 10-20 items and hallucinate imaginary elements to fill gaps.

3. Precise editing difficulties When you try to edit specific parts of an image, it might struggle with precision, either failing to make the requested change or altering too much.

4. Slow I feel guilty for commenting on speed for something this magical. But it can take one to two minutes to generate images, which is 10x as slow as image generation on Ideogram or other platforms.

If you’re interested in ethical considerations associated with AI image generation, watch this Ted Talk by Ed Newton Rex, founder of Fairly Trained, a non-profit that certifies generative AI companies that respect creators’ rights. Then watch a counterargument from artist Greg Lookerse.

Strong alternatives

* Ideogram also launched its new version (3.0) this week. It’s terrific, especially for abstract or metaphorical images, or for merging text with striking graphics like this. [See what I like about Ideogram.] Unlike ChatGPT, Ideogram has a menu for specifying an image’s dimensions and color palettes. And you can choose from four distinct image renderings.

Ideogram can’t accurately produce lengthy text inside images like ChatGPT, though, and it lacks other advanced capabilities. For now I'll continue paying $8/month for Ideogram, though the calculus is quickly changing.

* Adobe Firefly has a new standalone site. Its model is trained exclusively on material it obtained permission to use, making it a good choice for commercial projects. [See its ethics page].

* Reve is another great new AI image generation model that launched this week out of Silicon Valley. It renders typography well and abstract imagery like this. I like how you can modify images generated for you with a simple text prompt.

What image generation tools have you been experimenting with, and how? 👇

Special offer: reader discount on Letterly until April 1

I use Letterly to get past writer’s block, for journaling, and for ideation. It transcribes my rambling and reformats it into organized text. I use Letterly so much that its founders and I compiled a list of 50 ways to use the app.

Letterly’s founder is offering a lifetime deal specifically for Wonder Tools readers. Instead of paying for an annual $80 Letterly subscription, you can pay $149 once for lifetime access. You can get it through this unique link. The deal will be briefly active, just until April 1st at 11:59pm PST.

* Works on iPhone, Android, Mac, Web, iPad

* Unlimited recordings, transcriptions, and rewrites

* 90+ auto-recognized languages

* Record online or offline; widget for quick captures; screen-off recording

* 14-day money-back guarantee. You can cancel if it’s not useful for you.

* You won't find this deal elsewhere online.

* Transparency note: The link above is an affiliate link, so I get a small commission if you purchase through it to help fund Wonder Tools. I’m sharing this because I rely on Letterly and you might find it useful too.