Visual explanation has always been the difference between a concept that clicks and one that stays abstract. The problem for students and teachers was never knowing that a diagram would help — it was the hours required to make one. That constraint has largely dissolved, and the classrooms adapting fastest are not the ones with the biggest technology budgets. They are the ones that figured out which specific tasks these tools actually do well.
What Works Genuinely Well
Three educational uses have proven consistently reliable. Concept illustration comes first: generating a visual metaphor for an abstract idea — entropy as a shuffled deck, supply and demand as a physical balance — gives students a mental hook that pure text does not. Historical and scientific scene reconstruction comes second, useful for making a period or a process feel concrete, provided everyone understands it is an interpretation rather than a photograph. And presentation graphics come third, where a student preparing a seminar can produce coherent visuals in minutes rather than hunting through stock libraries.
The newest generation of models added a capability that matters enormously for education: readable text inside images. Labelled diagrams, annotated timelines, and titled charts were previously impossible to generate reliably. Current models handle short labels well enough that a first draft is genuinely usable, which opens up the single most common educational visual — the labelled diagram.
How Access Actually Works
Students often assume these tools require expensive subscriptions. In practice most educational tools reach image models through APIs and pay per generation, at costs measured in cents. A study group or a department can access GPT Image 2 via API through aggregation platforms that expose many models behind one account with published per-image pricing, which is why a small institution can now offer the same capability that a well-funded one does. The practical implication for a student: the barrier is knowing what to ask for, not what to pay.
The Academic Integrity Question
This deserves a direct answer rather than hand-waving. Using AI to generate an illustration for your own explanation is generally analogous to using a stock image or drawing a diagram — the intellectual work is your explanation, and the visual supports it. Using AI to fabricate data visualisations, experimental results, or anything presented as evidence is academic misconduct regardless of how the image was made. The line is not the tool; it is whether the image represents something you actually established.
Practical guidance most institutions have converged on: disclose AI-generated visuals in a caption, never generate anything that purports to be primary evidence, and check your specific course policy because they vary considerably. A one-line caption costs nothing and eliminates the entire category of problem.
What It Still Cannot Do
Accuracy in technical diagrams remains unreliable. A generated anatomical illustration will look convincing and place structures wrongly; a generated circuit diagram will look plausible and not work. For anything where correctness matters, these models produce style, not substance — use them for the visual language and verify every factual element against a real source. Students who learned this the hard way, usually in a graded submission, describe the lesson as memorable.
Used within those limits, the tools are genuinely valuable. The student who spends ten minutes generating and refining a visual metaphor for a difficult concept understands that concept better afterward — not because the AI explained it, but because articulating what the image should show forced them to be precise about what they actually understood.