From Candy Lands to Deepfakes: GPT-4o’s Image Generation and the New Parenting Challenge

Recently, OpenAI’s ChatGPT has demonstrated remarkable proficiency in a variety of tasks, from coding up accurate algorithms in a matter of seconds to blowing seemingly impossible high school math competitions out of the water. However, one particular release has excited users, and that is GPT-4o’s image generation. 

What can GPT-4o’s image generation do?

GPT-4o opens up many opportunities for children: It could create imaginative worlds, from a hyperrealistic candy land to a watercolor version of their favorite cartoon book, or bring their favorite stuffed animals to life by illustrating the toys in thousands of different art forms. It could provide countless resources for visual learners, making easily digestible graphs and infographics just one prompt away.

GPT-4o’s image generation brings in some new, exciting features, such as inpainting (image editing), high-fidelity generation, and interactive editing through natural language. The model has significantly improved in rendering human features — for example, hands and faces — in hyperrealistic images and videos. In addition, it can mimic artistic styles (Disney, for instance), express complex concepts, and generate photorealistic images. It can even generate diagrams, infographics, historical scenes, and scientific concepts that include words. This is a new development; before, ChatGPT had much trouble generating images with text (using random letters, spelling words wrong, and so on). 

How does GPT-4o’s image generation work?

Behind the scenes of creating these incredible images and videos, GPT-4o employs preexisting methods to achieve its impressive results. GPT-4o’s image generation uses a diffusion model trained on large datasets of images and text. Diffusion models learn to generate new data samples by gradually transforming random noise into complex distributions that resemble training data. Thus, in simpler terms, they learn to generate data similar to the data they are trained on. The “noise” is added to the data by simply feeding in random, meaningless information, ultimately creating a more robust training set. From this new data, GPT-4o learns how images relate to natural language, ultimately guiding it to exhibit strong visual fluency, to interpret the context given, and to use that understanding to create images. 

What do the advanced capabilities of GPT-4o’s image generation mean for parents?

Large multimodal models are able to create eerily realistic images, creating concerns for parents who have children who are especially vulnerable on the internet and social media. Here are some concerns and how they could be addressed.

Concern #1: What if my child uses this tool to create deepfakes and other inappropriate images?

This is a major concern ever since OpenAI released DALL·E, in 2021. Kent Place School has addressed it by having students study numerous cases on deepfakes, even one about the real experience of a teen girl in New Jersey. These discussions prompt students to reflect on the grave effects of misusing image-generative tools. The need for these conversations does not provide comfort to parents who are terrified of the capabilities GPT-4o’s image generation gives to malicious users. 

There are many ways for parents to keep their children safe. First, ChatGPT has guardrails in place as an effort to prevent the generation of inappropriate images. For example, if a user requests to make a hyperrealistic image of an actual person (by uploading an image of someone), making the individual look like a doll figurine, for example, it responds with something along the lines of: 

I couldn’t generate the image you requested because it violates our content policies. . . .The issue is that the request involved creating a realistic image of you based on your actual photo. Even though your idea was creative and fun, generating realistic or lifelike representations of real people (including yourself) from images you upload goes against our content policies. This is to protect privacy and ensure safety for all users. —ChatGPT, OpenAI, April 22, 2025, chat.openai.com

However, it is important to highlight that these are not always accurate in flagging potentially dangerous image generation. To combat this issue, parents could co-create images with their children to encourage ethical use and understanding of  AI image generation. In addition, ethical conversations regarding responsible use of image-generative tools and AI in general have proved to be useful in exposing the gravities of the creation of deepfakes. 

Concern #2: What if my child can’t differentiate between a real image and an AI-generated image, and is misled as a result?

This is a very real possibility for all ages — countless people are misled by AI-generated images posted on social media every day. To increase awareness, parents should give their children early media literacy education. One helpful exercise could be having kids look at AI-generated images and real images side by side, and asking them which they think is real. By helping children realize the capabilities of GPT-4o’s image generation through firsthand experience, they are more likely to approach other images more cautiously.

AI’s potential is limitless; daily, promising new capabilities are released, showing proficiencies in its applications in a variety of fields, among them medicine and finance. It has already shown remarkable accuracy and efficiency, exciting many people as it revolutionizes our world. However, the issue of safety must be emphasized. As the upcoming generation spends an increasing amount of time online, parents have an unavoidable duty to set their children up for success when navigating the digital world.

Image Citation: 

“A cat looking into a puddle of water on a street, but its reflection is that of a tiger, and both reflections are realistically distorted by ripples in the water” ChatGPT, GPT‑4, 2025. Prompt: “A cat looking into a puddle of water on a street, but its reflection is that of a tiger, and both reflections are realistically distorted by ripples in the water” AI-generated image.

Anna Bulto '26 and Skylar Li '26