In this guide 8 sections
An AI caricature from a photo needs one clear picture and a prompt that names the features to exaggerate and the ones to keep. We ran one such prompt on three image models on 2026-10-26, one try each, on the same reference photo. Our log marks the Nano Banana Pro result as a true caricature (2 credits). GPT Image 2 kept the closest likeness but exaggerated least (0.5 credits), and Seedream 5.0 Lite returned a generic cartoon (1 credit). That is one sample per model, not a ranking. Free routes and their vendor-stated limits follow the prompt and the fix section.
Part of the AI effects library.
One photo, three models: our results (tested 2026-10-26)
The input was the reference photo of our fictional, AI-generated adult test character: curly copper hair, round tortoiseshell glasses, freckles. All three models ran inside Higgsfield on a Plus-plan account, 3:4 frame, one try each. The job record lists our Nano Banana Pro run as nano_banana_2; the billing line reads “Nano Banana Pro”. We report both and conclude nothing.
Model (via Higgsfield) | Credits | Tries | What we saw |
|---|---|---|---|
Nano Banana Pro | 2 | 1 | Classic caricature: very big head on a tiny body, freckles and glasses strongly exaggerated, plain paper background. Hands drawn as stubs without fingers |
Seedream 5.0 Lite | 1 | 1 | Reads as a generic cartoon portrait: huge eyes with lashes, body cropped at the waist, not small |
GPT Image 2 | 0.5 | 1 | Big head on a small full body with jeans and sneakers, hands in pockets. Mildest exaggeration |
Total for the three: 3.5 credits (2 + 1 + 0.5). Every charge matched the quote we read before running. Here are the outputs.
This is a single sample per model. The settings were not identical either: GPT Image 2 ran at 1K low quality, Nano Banana Pro at 1K, Seedream at its default quality. Read the table as what happened on this prompt and this photo, not as a ranking.
The caricature-from-photo prompt and how to adapt it
This is the logged prompt, verbatim, used on all three models:
Tested prompt
The same text as in the recipe card above. Copy it from the card
To adapt it, change three things and leave the rest:
- Swap the subject. Replace “the woman in the reference photo” with a description that fits your photo. Use your own photo or a fictional character.
- Rewrite the feature list. Name two to four features that people notice first: hair, glasses, freckles, a smile. The prompt gives the model a list to push on instead of leaving “exaggerate” open.
- Keep the recognizability clause. “While she stays clearly recognizable” asks the model to keep the face. Seedream 5.0 Lite still went generic with it, so the clause is a request, not a guarantee.
All three results came back on the plain white background that the style sentence (ink lines, watercolor, white paper, plain background) asks for. We did not test the prompt without it.
This prompt is not the ChatGPT caricature prompt from the viral wave. That one, with a job and props, plus our no-memory run, lives on the AI caricature trend page. This page keeps the permanent recipe, a plain portrait caricature from photo input: one image in, one drawing out.
Which model kept the likeness while exaggerating
Our working definition for this test: a caricature pushes someone’s distinctive features while the person stays recognizable. A cartoon portrait restyles the face into a generic one, and exaggerated features are what separate the two. Judged by that, the three outputs split like this.
- Nano Banana Pro pushed hardest. The head is oversized, the freckles and glasses are enlarged, and the output still reads as the same person. Our log marks this one alone of the three as “true caricature, likeness kept”.
- Seedream 5.0 Lite kept hair, glasses and jacket, but the face went generic: huge cartoon eyes, lashes, a body that is not small. By our definition this is a cartoon, not a caricature.
- GPT Image 2 via Higgsfield kept the closest likeness of the three and exaggerated the least. The big head and small body are there, but the features are pushed mildly.
The pattern on this one run: the more a model exaggerated, the further it sat from a strict portrait, and the model that stayed closest to the face pushed least. That is an observation of one photo, one prompt and one try each. If you want more exaggeration from the closest-likeness model, a stronger feature list in the prompt is the first thing to change. We did not test that.
Fixing the face, hands or background with one follow-up edit
We planned up to three targeted edits, one per defect: face, hands, background. Only one defect appeared. Nano Banana Pro drew the two tiny hands as stubs without fingers. The face kept the likeness and the background was the plain paper we asked for, so those two fixes were “no fix needed” and we ran no edits for them.
The edit used the first output as the image reference and this prompt, verbatim:
Tested prompt
Keep this caricature drawing exactly the same: same face, hair, glasses, freckles, jacket, pose, style and white paper background. Only redraw her two small hands so that each hand has clearly drawn fingers.
It ran once on Nano Banana Pro and charged 2 credits. Result: both hands now have drawn fingers, and face, hair, jacket, pose and paper held. Two side effects showed: small differences in the curls and freckle pattern, and the hands came out slightly large for the tiny body.
How to repeat it: start the edit with “keep this exactly the same”, list what must not change, then name the one thing to redraw. One defect per edit. Fixing the hands cost as much as the original image, so the full test of three models plus this fix was 5.5 credits.
Free caricature makers: limits and sign-up (vendor-stated, dated)
We ran the test on a paid Higgsfield Plus account, so it does not show what any free account gets. Start free: ChatGPT and Gemini both accept an uploaded photo with an image request. Their free limits are not verified as of 2026-10-26, and this prompt on either one is not tested. Adobe Firefly, Canva and Pixelbin pages also rank for “ai caricature” (our Google US snapshot, 2026-10-20), and “free caricature app” lists name them; we have not checked their limits and list none. What Higgsfield itself gives free is on the free tier page.
We checked the free-offer pages of these two tools only. Figures are the vendors' own wording, not our measurements.
Tool | Free offer as stated | Sign-up | Checked |
|---|---|---|---|
Dreamina (dreamina.capcut.com) | Free caricature maker "with 150 everyday credits"; its page says "no credit card required" | Buttons lead to a sign-in page | 2026-10-20 |
GoEnhance (goenhance.ai) | "Generate caricature-style images for free with limited tokens"; upgrade for advanced features | Not stated on the page | 2026-10-20 |
Neither page says how many caricatures the free allowance covers. Watermark and export terms: not stated on those pages as of 2026-10-26. Check them before you rely on a result for a gift or a profile picture.
Our tests ran on Higgsfield. We may earn a commission if you subscribe through the link below; it does not change what we test. Details are in our affiliate disclosure. To recreate the three runs, open Higgsfield, choose Image, pick a model and add your photo as the reference: Open Higgsfield (commission link).
Paying makes sense if you want the same prompt on several models at 0.5 to 2 credits a try (our charge, 2026-10-26). For one AI caricature as a gift or a profile picture, try a free route first.
Before you post: your photo, other people
Everything here is AI-generated: the test character is fictional and an adult, and every image carries that label. We publish no how-tos for other people’s faces; see our content policy. Use your own photo.
A caricature exaggerates a face, and that can hurt. Get the consent of any adult in a photo before you upload it or post the result, and never use a photo of a child. How long a tool keeps an uploaded photo is set in its own privacy terms, which we have not checked for the tools above. Read them before you upload a photo of yourself.
Quick answers
How do I get AI to make a caricature of me?
Upload one clear photo to an image model that accepts a reference, paste the prompt above, and name the features to exaggerate. We ran it on Nano Banana Pro, Seedream 5.0 Lite and GPT Image 2 via Higgsfield on 2026-10-26 (2, 1 and 0.5 credits, one try each).
How to convert a photo to a caricature for free?
Try ChatGPT or Gemini first; their free limits are not verified as of 2026-10-26. Dreamina states 150 everyday credits and GoEnhance states limited free tokens (both checked 2026-10-20).
Caricature vs cartoon: what is the difference?
A caricature exaggerates your distinctive features and stays recognizable. A cartoon restyles you into a generic face. In our test, Seedream 5.0 Lite returned the cartoon, Nano Banana Pro the caricature.
Sources checked
- Our test: How we test and the fictional test character. Runs H11a, H11b, H11c and H11-fix1, 2026-10-26, Higgsfield Plus account via Higgsfield MCP; prompts, job records and credit charges are from our session log.
- Dreamina, Caricature generator from photo: “150 everyday credits”, “No credit card required”. Checked 2026-10-20; the wording was unchanged on 2026-10-26.
- GoEnhance, AI caricature generator: free “with limited tokens”, “Upgrade for advanced features”. Checked 2026-10-20; the wording was unchanged on 2026-10-26.
- Higgsfield free-tier terms: our free tier page, linked in the free section. We did not check official model pages for Nano Banana Pro, Seedream 5.0 Lite or GPT Image 2, so this page makes no vendor claims about them.
- The caricature trend and the no-memory job-prompt test: AI caricature trend.
- Search results: our Google US snapshot for “ai caricature”, 2026-10-20.
