Straight answer: ChatGPT can turn a photo into a cartoon caricature if you upload a clear, front-on image and give it a named art style plus specific facial features to exaggerate, but you should expect five to ten attempts before you get one that looks like the person. It is free to try in a limited way and costs 20 US dollars a month for ChatGPT Plus if you want to do it and often.
What I did with this, on a real client project
Last month a client running a small accountancy firm in Reading wanted cartoon avatars for the "meet the team" page on her website. Six people, six headshots, a budget of zero because the illustrator quote had come back at 80 pounds per person. I said give me an hour and let's try ChatGPT.
It took closer to ninety minutes and fourteen separate image generations to get six usable cartoons. Two of the six photos worked on the first or second try. One person, a man with a fairly distinctive gap in his front teeth and a shaved head, took nine attempts because ChatGPT kept giving him a full head of hair and straight teeth. I had to write out, in plain words, "keep the gap between the front teeth, keep the head shaved, do not add hair." Once I named the specific feature instead of just saying "keep it recognisable," it got it right.
That's the bit nobody tells you upfront: ChatGPT's image tool has a habit of quietly beautifying people. It smooths skin, straightens teeth, slims faces, and de-ages people by what looks like five to ten years unless you fight it on every single detail. If your caricature needs to look like the actual human and not a slightly nicer stranger who's wearing their clothes, you have to be blunt about the features you want kept.
What you need before you start
- A clear, well-lit photo taken face-on, not at an angle. Side profiles and photos taken from below confuse it badly.
- Either a free ChatGPT account, which gives you a small number of image generations per day (roughly two to three before you hit a limit), or ChatGPT Plus at 20 dollars a month (about 16 pounds in the UK), which gives you far more headroom and faster generation.
- A named art style. "Make this a cartoon" gives you a generic, forgettable result. "Make this a 1990s Simpsons-style cartoon" gives you something specific.
- Five minutes you're willing to spend going back and forth, because the first result is rarely the final one.
The step-by-step process
Here's the actual sequence I use now, after that Reading job taught me the shortcuts:
- Step 1: Open ChatGPT (app or browser) and upload the photo directly into the chat.
- Step 2: Name a specific style in your first prompt. Don't say "cartoon," say "Pixar-style 3D character" or "New Yorker magazine pen-and-ink caricature" or "Simpsons character."
- Step 3: List two or three exact features to keep unchanged: glasses, hairstyle, a beard, a specific hair colour, a nose shape. This is the single biggest lever for getting a likeness that resembles the person.
- Step 4: Say what to exaggerate, if it's a true caricature rather than a stylised portrait. Caricature, technically, means exaggerating one or two features for comic effect, that's the actual definition according to Wikipedia's entry on caricature, and if you skip this step ChatGPT just gives you a flattering cartoon portrait instead.
- Step 5: Ask for the background to be simple, a plain colour or none at all, so you can use it on a website, in an email signature, or on a slide without a messy background fighting for attention.
- Step 6: Look at what comes back and correct it in plain, specific language. "Keep the glasses but make them black-framed like the photo," not "make it look more like me."
- Step 7: Once you're happy, ask ChatGPT to generate it again at a higher resolution or a square crop if you need it for a specific platform, since the default output is usually 1024 by 1024 pixels.
Prompts that work
These are the exact prompt structures I've used and had good results with, word for word:
Pixar or animated-movie style:
"Turn this photo into a 3D Pixar-style cartoon character. Keep my exact hairstyle, hair colour and glasses recognisable. Give it warm, soft animated-movie lighting and slightly rounded features, but don't change my face shape or skin tone. Plain light grey background, no text."
New Yorker magazine style caricature:
"Create a black and white pen-and-ink caricature of this photo in the style of a New Yorker magazine illustration. Slightly exaggerate my eyebrows and nose for comic effect but keep the rest recognisably accurate to the photo. Add cross-hatching shading. White background."
Simpsons style:
"Redraw this photo as a Simpsons character. Yellow skin, my exact hairstyle and hair colour, keep the glasses from the photo, keep the beard if there is one in the photo. Standing against a plain pale blue background, upper body only."
Watercolour cartoon, good for a softer brand feel:
"Turn this photo into a loose watercolour cartoon portrait. Keep my face shape, hairstyle and glasses true to the photo. Use warm, muted colours, visible brush texture, and a plain cream background. Don't make it look like an anime character."
Notice none of these say "make it look like me" on its own. That phrase is too vague for the model to act on. Naming the exact features, the exact style, and what not to do is what moves the result.
The uncomfortable bit nobody selling "caricature prompt packs" mentions
There's a small industry right now of people selling "50 ChatGPT caricature prompts" for a few pounds on Etsy and Gumroad. Most of them are one or two sentences you could write yourself in thirty seconds once you know the structure above. I'm not against paying for convenience, but you're not buying secret prompts, you're buying a list someone typed up once.
The bigger issue is likeness accuracy. In my own testing across roughly 40 images for different clients over the past few months, I'd say the model nails a genuine, recognisable likeness on the first or second attempt about 4 times out of 10. The rest need feature-level correction, and a handful never quite get there no matter how specific you are, usually with older faces, faces with asymmetry, or faces from lower-quality source photos. If someone tells you it's a one-prompt, one-click job every time, they haven't done it more than a few times themselves.
There's also a sameness problem worth being honest about. Because so many small businesses are now using the same three or four style prompts (Pixar, Simpsons, New Yorker), if you put your cartoon team avatar next to a competitor's, there's a real chance they look like they came from the same template, because they did. If you want your business's visual identity to stand out rather than blend into every other "meet the team" page using ChatGPT this year, either push for a unusual style prompt (Ukiyo-e woodblock, Bauhaus poster, 1960s British comic strip) or accept that this is a fast, cheap solution rather than a distinctive one.
One more thing worth knowing before you build anything commercial on this: ChatGPT will sometimes refuse to generate an image based on a real, identifiable person's photo, particularly if the person looks like a public figure or if the request reads as trying to mock or defame someone. This isn't a bug, it's a policy guardrail. If you're doing this for a business asset like a team page, book cover, or merchandise, it's worth getting a quick written OK from the person in the photo anyway, since likeness and consent questions don't disappear just because a cartoon version was easy to make.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Where this earns its keep
For all the caveats, there are real use cases where this is worth the ninety minutes:
- Team pages and about-us sections where a consistent, friendly cartoon style beats mismatched stock headshots.
- Social media avatars for founders who don't want to keep updating a real photo every time they change their hair or their glasses.
- Quick internal fun, birthday cards, leaving cards, slide decks that need a bit of personality.
- Speaker bios and event programmes where a stylised image looks more polished than a phone photo, without the cost of hiring an illustrator for one person.
What it's not great for yet: printed merchandise that needs a sharp, high-resolution vector-style output, or anything where brand consistency across dozens of images matters more than speed, since getting the exact same style twice in a row still takes real fiddling.
A quick note on cost and time, honestly
If you're doing this for one photo, the free tier is fine, budget for three or four attempts before you hit your daily limit and have to wait or upgrade. If you're doing this for a team of six or more, Plus at 20 dollars (roughly 16 pounds) a month pays for itself against a single illustrator quote, and you'll likely spend somewhere between ten and twenty minutes per person once you know the prompt structure, rather than the ninety minutes it took me the first time I did it cold.
To write your own, try the free Free ChatGPT Prompt Generator: Build a Better Prompt in 30 Seconds.
Related guides live in the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.
Related reading: best chatgpt photo editor.
Related: dating: guidelines and how to pitch.
Frequently asked questions
Is turning a photo into a cartoon with ChatGPT free?
Yes, up to a point. The free tier gives you a limited number of image generations per day, usually around two or three, which is enough for a single photo if you're patient. For a whole team or repeated use, ChatGPT Plus at 20 dollars a month gives you far more attempts and faster results.
Why doesn't the ChatGPT cartoon look like the person in the photo?
The model tends to smooth skin, straighten teeth, and slim faces by default. To fix this, name the specific features you want kept exactly as they are in the photo, such as glasses, a hairline, a beard, or a gap in the teeth, rather than saying "keep it recognisable," which is too vague for it to act on.
Will ChatGPT refuse to caricature a photo of a real person?
Sometimes, particularly if the person resembles a public figure or the request reads as mocking someone. For business use, it's also worth getting permission from the person in the photo before you use their cartoon on a website or in marketing, separate from any policy question.
What's the best ChatGPT prompt for a caricature that looks like a real caricature, not just a cartoon portrait?
Name the style directly, such as "New Yorker magazine style pen-and-ink caricature," and explicitly ask it to exaggerate one or two features for comic effect, such as the eyebrows or nose, while keeping everything else accurate to the photo. Without that instruction, you'll get a flattering cartoon portrait rather than a true caricature.
