Yes, the new ChatGPT image generator is an upgrade. But teachers beware.
If you missed it, Open AI just released a GPT-4o image generator. Rolled out initially to both free and paid ChatGPT users as the default image generator, it received so much traffic on its first day that circuits were “melting” and OpenAI temporarily suspended access to free users.
According to OpenAI, the new GPT‑4o image generator “excels at accurately rendering text, precisely following prompts, and leveraging 4o’s inherent knowledge base and chat context.” In more practical terms, the images it produces are more realistic, easier to prompt conversationally, and more diverse. To wit, the new GPT‑4o image generator can create an impressive photorealistic image of two witches in their 20s reading a fictitious street sign, a four-panel comic strip, and a boarding pass — all without any apparent errors.
Previously, ChatGPT created images using its DALL-E 3 image model. However, those images were far from photorealistic, frequently inaccurate, contained jibberish text, and often required multiple prompts to achieve the desired result. Now, all image generation is handled by “omnimodal” GPT-4o — a model that can create any data, including text, image, audio, and video by drawing from its extensive knowledge base.
But should you use it?
The new GPT‑4o image generator has received enthusiastic reviews and garnered widespread interest. But, after taking a spin, I wouldn’t recommend teachers abandon Canva or Adobe Firefly anytime soon.
Here’s why:
So, I’d venture that most K12 educators want to create images for these general instructional uses:
for presentations
to accompany lesson plans
for posters
as story starters
for visual flashcards
to incite conversations
as an illustrated guide
for classroom illustrations
They’d also want the images to be age-appropriate, relevant, clear, and editable.
The question then is whether the GPT-4o image generator can meet these needs — quickly and accurately.
The answer: Somewhat.
Let’s start with a positive. ChatGPT 4o allows educators to generate images and lesson content in one conversation. It claims you can say, "Give me an elementary-school lesson plan about the water cycle and a cartoon-style diagram to go with it," and it will deliver both—no need to jump between platforms.
But let’s look at the output when we try this example.
First, here is the lesson plan it generated:
🧪 Lesson Plan: The Water Cycle
Grade Level: 3rd–5th Grade
Subject: Science
Topic: The Water Cycle
Time: 45 minutes
🎯 Objectives
Students will be able to:
Identify and explain the main stages of the water cycle: evaporation, condensation, and precipitation.
Illustrate the cycle using visuals and appropriate vocabulary.
🧰 Materials
Projector or printed water cycle diagram
Poster paper or science notebooks
Markers or colored pencils
🧠 Introduction (5–10 min)
Begin with a simple question: “Where does rain come from?”
Show the cartoon-style diagram of the water cycle.
Introduce vocabulary: evaporation, condensation, precipitation.
📝 Activities (25–30 min)
Teacher Explanation:
Use the diagram to walk through the water cycle. Emphasize how water moves and changes states.Student Illustration Activity:
Students draw their own version of the water cycle and label each stage.
Optional: Create a story or comic strip showing the journey of a water droplet.Pair Share:
Students explain their diagrams to a partner using vocabulary words.
✅ Assessment / Check for Understanding (5–10 min)
Ask students to verbally walk through the cycle or point to the correct stage when prompted.
Have a few students present their drawings.
🌟 Extension (Optional)
Create a classroom mural of the water cycle with contributions from each student.
Tie in water conservation or climate awareness.
Basic, but a satisfactory framework.
Next, here’s the image it created (after about a minute):
Notice that the lower “evaporation” is spelled incorrectly. And its insertion creates a level of confusion. I asked ChatGPT to correct the error, but it did not; it simply recreated the same image.
A digestive system lesson plan
Next, I asked ChatGPT to “Give me a middle school lesson plan about the digestive system and an age-appropriate image to go with it.” Here is what it created:
It didn’t create a separate, text-based lesson plan. Instead, it tried to cram the lesson plan into the image. Not surprisingly, the text is cut short; it simply cannot accommodate all the elements in the lesson plan.
Historical Events in Boston During the Revolutionary War
Next, I tried to create a poster: “Create a poster for a middle school social studies classroom on historical events in Boston during the War for Independence.”
Here is what it created after about two minutes:
Obviously, it didn’t complete the task. So, I asked it to complete the task. About two minutes later, it created a third of the image and halted.
Yet, when I returned to ChatGPT sometime later (15 minutes?) this is what I saw:
It’s useful as a unit poster, though there is a slight error (“-776”). Surprisingly, ChatGPT generated three versions of the same poster with stylistic variations.
Photo-realistic Colombian street market
Finally, I asked ChatGPT the following: “Create a photo-realistic cultural scene in a Colombian street market for use in a Spanish II class.”
Here is what I saw for the first five+ minutes:
Yes, a blank image. So, I asked ChatGPT to complete the image generation. It provided me with a link to download the image. But the link didn’t work, saying, “The session expired.” I told ChatGPT as much, and it provided me with a new link to download the image. Again, the link didn’t work and I received the same error message. Ugh.
But Wait!
I abandoned the attempt and continued to write this post. But, lo and behold, when I returned to ChatGPT about 10 minutes later, this was waiting for me:
As my Colombian wife commented: “Que hermoso. Se ve muy real.” (“How beautiful. It looks quite real.") The image could be a great conversation starter and a chance for students to build or reinforce vocabulary.
In partnership with Essay Grader
Essay Grader is a straightforward, powerful tool that uses AI to take the load off your shoulders and will completely change the way you approach grading!
Every feature puts teachers first, with a focus on simplifying repetitive tasks like grading to improve productivity - all the while letting teachers be fully in control.
Streamline grading, provide detailed, actionable feedback to your students, and maintain complete control over your assessments - all in one intuitive interface.
Reduce grading time by 80%. Grade your entire class's essays in 2 minutes or less and deliver high-quality, specific feedback to your students.
Ensure fair essay grading by removing bias and aligning with state standards like Texas STAAR, Florida BEST and California CCSS.
Join Essay Grader and its ever-expanding network of 60,000+ educators, and you’ll connect with others who share your passion and commitment.
More struggles
By now, I’m sure you get the picture (forgive the pun): You’ll need patience. The GPT-4o image generator is not entirely ready for K12 classrooms. At times, it displays a promising combination of convenience, speed, accuracy, and flexibility. At other times it’s just plain clunky.
OpenAI openly acknowledges limitations. For example, it says the GPT‑4o image generator “struggles to accurately render more than 10-20 distinct concepts at a time.” So, if you’re hoping to create a detailed visual guide — such as the periodic table — you should look elsewhere.
In addition, it can struggle with “precise graphing.” I tried some “not-so-precise graphing,” and it struggled to depict the data accurately. So, complex math visuals are obviously out of the question.
The GPT-4o image generator also has difficulty rendering non-Latin languages, as “the characters can be inaccurate or hallucinated.” So, Chinese teachers will likely have little use for it.
When Might Canva or Adobe Firefly Be Better?
The GPT-4o image generator will undoubtedly improve, but there are clearly great reasons to opt for Canva and Adobe Firefly in many classroom contexts:
Polished, Branded Designs (Canva)
If you're designing school flyers, newsletters, posters, or social media graphics, Canva shines. It offers:
Templates for professional-looking materials.
Easy drag-and-drop design tools.
Collaboration features for teams or students.
And Canva for Education is free for eligible teachers and students all over the world.
Artistic Style & Photo Realism (Firefly)
Adobe Firefly is ideal when:
You want stylized, artistic, or photorealistic images.
You're working with older students on graphic design or digital art projects.
You need integration with Adobe tools (like Photoshop or Illustrator).
Student Projects
In addition, Canva and Firefly both have student-safe environments and class management tools. They may be better suited for students to create their own work, while ChatGPT’s image generator is geared more towards teacher-created content.
Have you tried ChatGPT’s new image generator? Any comments, or suggestions? Please provide them below.
AI in Edu: News, Tools & Views is sent to all subscribers on Fridays.
Find more AI tools, views, and how-to’s at tomdaccord.com
Find my book AI Tools & Uses: A Practical Guide for Teachers at Amazon
*Learn more about Essay Grader
Great piece. I was under the impression that the image generator in Canva uses Dalle 2 (or 3?).
I use Canva oder Dalle-3 in ChatGPT for images.
The way I see it, the software gods have to master 3 things before teachers can fully use 1) character consistency (for storyboards etc) 2) anatomical accuracy in body parts but also organs so that they can be correctly labeled and 3) text / writing / spelling. If they have already achieved no 3 then yay! We can expect this to be standard with all tools soon.
What do you think?