ChatGPT Caricature Prompts: Style-Locked Portrait Kits

A ChatGPT caricature prompt turns a face into an exaggerated portrait in ChatGPT Image (GPT Image). The useful move is a style-locked kit: fixed exaggeration rules, fixed art medium, and bracket slots for subject cues you swap each run. Upload your own selfie or describe a generic sitter, paste a kit, iterate in thread. You leave with five paste-ready kits, a selfie workflow, mid-2026 GPT Image notes, and a soft path to PromptMake /image for reference photo text. Caricature kits only; not a generic viral trend roundup.

What a ChatGPT caricature is (and who uses it)

A caricature portrait exaggerates selected facial features while the sitter stays recognizable. ChatGPT Image reads your upload or your text brief and renders the look in one thread. You can ask for a softer editorial exaggeration or a bold street-art push. The model responds to how you name the exaggeration budget and the art medium.

Searchers who type "chatgpt caricature" usually want paste prompts that work on their own photo. They also want the look to hold when they run a second portrait for a teammate or a client-approved subject. That is why style-locked kits beat one-off viral strings. The kit names what stays fixed across every render.

This article differs from our trending ChatGPT image prompts guide, which teaches slot maps for any viral look (clay product, disposable night street, chrome hero). Here the medium is caricature portrait grammar: feature emphasis, line weight, color blocks, and likeness limits. For general describe, recreate, and restyle jobs on any still, see ChatGPT prompts for photos. For mode picking on portrait uploads (Recreate, Relight, Restyle), see the portrait photo to prompt guide.

Strong fit for:

• Creators who want a repeatable caricature look for profile art or channel avatars
• Designers who must deliver caricature variants from an approved reference selfie
• Teams who need one exaggeration style across three sitters without rewriting from scratch
• Anyone who exports a locked caricature brief into Midjourney v7 or FLUX after ChatGPT nails the look

Skip this page if you need photoreal headshots with no exaggeration. The headshot prompt guide covers that lane. Skip celebrity or public-figure likeness targets. Use your own photo, a consented subject, or a generic character brief.

Anatomy of a style-locked caricature kit

Every usable ChatGPT caricature kit splits into a fixed half and a variable half. The fixed half is the art contract: exaggeration level, line or brush grammar, palette, lighting, background, and hard constraints (single subject, no text, chest-up crop). The variable half is what you swap: face cues from an upload, wardrobe color, expression, optional prop. If you paste a full viral string built for someone else's jawline, leftover nouns fight your sitter. Strip first. Slot second.

Treat exaggeration as a dial you name in the fixed half. "Moderate: emphasize nose and brow, keep ears natural" beats "funny face." Treat medium as one noun: ink editorial, gouache editorial, vector flat, 3D vinyl toy. Two media in one line muddies the series. Write fixed and variable lists before you generate. The subsections map the slots caricature kits use most often.

Fixed slots: exaggeration, medium, light, constraints

Exaggeration level is the caricature dial. Name what grows and what stays normal. Common patterns: larger eyes with smaller mouth, elongated chin, wider smile, bigger hair volume, stronger brow ridge. Pick one or two emphasis points per kit. Three emphases at max strength often collapse into a generic cartoon.

Medium locks the render family. Ink editorial caricature uses clean contour lines and flat color blocks. Gouache editorial adds visible brush edges and paper texture. Vector flat uses bold shapes and limited palette. 3D vinyl toy uses smooth plastic skin and studio product light. Light and grade follow the medium: soft overhead for toy, side key for ink, warm paper bounce for gouache.

Constraints stop the usual failures: single subject, no readable text, no extra limbs, plain or simple gradient background, chest-up or head-and-shoulders crop, natural skin texture words where the medium allows. Cut empty praise ("masterpiece," "8K"). They burn tokens and do not lock a look.

Variable slots: subject cues, expression, wardrobe, props

Subject cues come from your upload or a short text description: round glasses, short curly hair, light stubble, freckles across the nose. Write cues as visible facts, not identity claims. Expression and wardrobe sit in the variable half: warm smile, neutral mouth, navy crew neck, red scarf. Props stay optional. One prop per kit keeps the face readable.

Use brackets in the shell: [subject cues], [expression], [wardrobe], [optional prop]. Fill brackets in a second pass after the fixed half is solid. Never leave a proper noun from a viral original inside a variable slot. That is how someone else's hat color lands on your sitter.

How to run caricatures in ChatGPT Image

ChatGPT Image (GPT Image) is the native image generator inside ChatGPT. You upload a reference or describe the sitter, paste a kit or a fill instruction, and iterate in the same thread. Follow-ups like "more brow emphasis" or "flatter background" are cheap when the fixed half already names the medium. Fast chat often routes to GPT-5.5 Instant. Dense style breakdowns on a tricky reference may read cleaner on GPT-5.6 Sol, Terra, or Luna when your plan exposes them.

Work in three passes when you start from zero. Pass one: pick a kit from the list below or build a fixed shell. Pass two: upload a consented reference or fill text cues. Pass three: one edit round on exaggeration before you export or copy text for another model. Plan fifteen minutes for the first successful caricature. Later runs shrink once your kit library holds.

From a reference selfie

Upload a clear face-forward selfie or a cropped head-and-shoulders still. Even light helps. Harsh shadow under the nose makes exaggeration guess wrong. Ask ChatGPT to read visible cues only, then apply your fixed kit.

Paste instruction with upload:

Using the uploaded photo as reference, generate a caricature portrait. Apply this fixed style kit exactly: [paste fixed half from a kit below]. Use visible features from the upload only. Do not invent props. Do not change the art medium. Single subject, chest-up, plain background. Output one image.

After the first render, steer with short edits: "reduce eye size ten percent," "stronger chin emphasis only," "cooler palette, same line weight." Change one exaggeration axis per edit. Parallel edits hide which dial broke the likeness read.

Consent and likeness limits: use your own face, a subject who agreed, or a fictional character brief. Do not target celebrities, politicians, or living artists by name to mimic their likeness. Platforms restrict that use. Ethical briefs also produce cleaner iterations because the model is not fighting policy guardrails.

From text-only prompts

Text-only caricatures work when you have no upload or when policy blocks a face file. Describe visible cues in plain nouns: "middle-aged man, rectangular glasses, gray buzz cut, light stubble, navy shirt." Attach the fixed kit. Avoid naming real people.

Text-only paste:

Generate a caricature portrait from this brief. Subject cues: [list visible features]. Fixed style kit: [paste fixed half]. Expression: [warm smile / neutral]. Wardrobe: [color and garment]. Single subject, chest-up, plain background. One image only.

Text-only runs drift more than upload runs. Plan two to three edit rounds on exaggeration. Save the winning fixed half under a kit name ("ink editorial moderate v2") so the next subject starts from a tested shell.

Five style-locked caricature kits to paste

These kits are starting points. Fill variable brackets per sitter. Test each kit on two different subjects (your selfie plus a text-only brief) before you file it. If the look holds, promote the kit to your library. If one subject collapses, tighten medium and exaggeration words, then retest.

Kit 1, ink editorial moderate:

Fixed: editorial ink caricature, clean black contour lines, flat color blocks, moderate exaggeration emphasizing [primary feature] and [secondary feature], soft side key from camera left, muted warm palette, plain off-white background, chest-up, single subject, no text.

Variables: [subject cues], [expression], [wardrobe].

Kit 2, gouache magazine:

Fixed: gouache editorial caricature on textured paper, visible brush edges, moderate-to-strong exaggeration on [primary feature], gentle top light, warm paper bounce, simplified background wash, head and shoulders, single subject, no logos.

Variables: [subject cues], [expression], [wardrobe color].

Kit 3, vector flat avatar:

Fixed: flat vector caricature portrait, bold shapes, limited four-color palette, strong exaggeration on [primary feature], even graphic lighting, no gradients on skin, plain solid background, square-friendly chest-up, single subject, no text.

Variables: [subject cues], [expression], [accent color on wardrobe].

Kit 4, 3D vinyl toy:

Fixed: 3D vinyl toy caricature figurine, smooth plastic skin, moderate exaggeration on [primary feature], soft product studio light, subtle floor shadow, neutral gray backdrop, three-quarter chest-up, single subject, no packaging text.

Variables: [subject cues], [expression], [shirt color].

Kit 5, street chalk board:

Fixed: street chalk caricature on dark board texture, dusty pigment edges, moderate exaggeration on [primary feature] and smile width, warm side light, visible chalk stroke direction, bust crop, single subject, no written words in frame.

Variables: [subject cues], [expression], [hat or glasses if visible in reference].

Fill instruction you can reuse after you pick a kit:

Fill this caricature kit for a new subject. Keep every fixed style clause. Replace bracket variables only. Subject cues: [list]. Expression: [ ]. Wardrobe: [ ]. Output one ChatGPT Image prompt line, then generate one image.

Export note: ChatGPT Image holds the look in thread. When you need Midjourney v7 or FLUX outside chat, copy the fixed half plus filled variables into a recreate brief. Midjourney wants short phrases plus --ar 1:1 or --ar 4:5 for portrait canvases and --style raw when the kit is photographic-leaning. FLUX wants natural-language material words (vinyl sheen, chalk dust, gouache bloom). Keep a dialect column in your notes.

Common mistakes with ChatGPT caricature prompts

Mistake 1: No exaggeration budget. "Funny caricature" returns random cartoon proportions. Name which features grow and how strong.

Mistake 2: Stacking two art media. Ink plus 3D toy plus watercolor in one line fights itself. One medium per kit.

Mistake 3: Chasing celebrity likeness. Prompts that target a famous face waste runs and violate platform rules. Use consented references or generic cues.

Mistake 4: Skipping the two-subject test. A kit that only works on your selfie stays a one-off. Run a text-only brief as the second subject.

Mistake 5: Ten parallel edits. Change one exaggeration axis per follow-up (chin, eyes, smile, palette). Mixed edits hide the fix.

Mistake 6: Photoreal language on a stylized kit. "Ultra realistic skin pores" on a vector flat kit pulls the render off brief. Match texture words to the named medium.

Mistake 7: Wrong export dialect. Midjourney flags pasted into FLUX waste a generation. Label the target model before you copy out of chat.

Mistake 8: Uploads without consent. A friend's face in a public chat thread is a policy and trust problem. Get permission or use your own still.

GPT Image and multi-model notes for mid-2026

GPT Image / ChatGPT Image stays the default conversational printer for caricature work in August 2026. Upload, kit, iterate, download. That loop is the trend driver behind "chatgpt caricature" search volume. Chat defaults often land on GPT-5.5 Instant for quick fills. Step to GPT-5.6 Sol, Terra, or Luna when you need a careful strip of a long social paste into fixed slots.

Claude Fable 5 and Claude Opus 5 handle kit wording and bracket fills if you move the text step out of OpenAI chat. Gemini 3.5 Flash is fast for short style notes; Gemini 3.1 Pro helps when a viral paste is long and contradictory before you template it.

Export targets still diverge after ChatGPT nails the look. Midjourney v7 (and V8 Alpha where you have access) wants concise visual phrases, aspect flags, and optional --sref when a reference still holds style after text is locked. Name FLUX.1.x or Flux 2 for your stack; both prefer flowing material language. Ideogram v3 fits poster caricatures only when lettering is the point. Leonardo and SDXL accept denser tags and short negatives in many UIs.

Reasoning-class models follow goal, constraints, and output shape. Skip "think step by step" theater on Sol-class runs. Instant and Flash tiers follow role, task, format, and one short filled example when the shell shape must match a prior good kit.

When PromptMake /image helps with a reference selfie

ChatGPT Image wins while you learn the caricature look and iterate exaggeration in thread. You already live in the upload plus edit loop. Stay there for first fills and kit tuning.

PromptMake /image wins when the look lives in a reference selfie and you need calibrated text for Midjourney, FLUX, DALL·E, Stable Diffusion, or Leonardo without hand-rewriting dialect rules. Pick a goal mode, generate once, edit the draft, paste into your external generator.

Pair Change Style with a selfie when you hold pose and crop but want the caricature medium named in prose (ink editorial, gouache, vector flat). Pair Recreate Exactly when the reference already shows the exaggeration level you want captured in words. Pair Adjust Lighting when the caricature kit's light grammar is the main gap. Pair Create Variation when you want sibling caricature briefs from one sitting for exploration.

Soft start: https://promptmake.net/image. Guests get about three image runs per day; free registration raises that to about five. Image and text quotas on PromptMake are separate, so caricature photo work does not spend your /text budget.

Privacy note: client faces, unreleased campaigns, and sensitive locations may not belong in a public chat upload. Use a tool with clear retention rules, or keep files local, when policy requires it. Redact background details you do not need in the prompt.

FAQ

These questions cover what people search after their first ChatGPT caricature run. Topics include kit structure, selfie uploads, exaggeration control, GPT Image in 2026, PromptMake /image for reference photos, export to Midjourney and FLUX, and likeness limits. Each answer stays short so you can act in the same session.

What is a ChatGPT caricature prompt?

It is a structured image brief you paste into ChatGPT Image to render an exaggerated but readable portrait. Strong prompts lock the art medium, exaggeration rules, lighting, and background in a fixed half, then swap subject cues in a variable half. Kits beat one-off viral strings because the style survives when you change sitters.

How do I make a caricature of myself in ChatGPT?

Upload a clear consented selfie, pick a style-locked kit from this page, and paste the fixed half plus your visible cues. Generate once, then edit one exaggeration axis per follow-up (chin, eyes, smile, palette). Save the fixed half under a kit name when the look holds on a second subject test.

What is the best exaggeration level for ChatGPT caricature prompts?

Name moderate as one or two emphasized features at readable strength. Editorial ink and gouache kits often use moderate emphasis. Vector avatars and 3D toy kits tolerate stronger single-feature push. Avoid maxing every feature at once; the face reads as a generic cartoon and loses likeness.

Can I use ChatGPT caricature prompts without uploading a photo?

Yes. Describe visible cues in plain nouns (glasses shape, hair length, wardrobe color) and attach a fixed kit. Text-only runs drift more than upload runs, so plan two or three edit rounds on exaggeration. Do not name real people; use generic or fictional subject language.

Is it okay to caricature celebrities with ChatGPT?

No for most production use. Platforms restrict public-figure likeness generation, and the prompts fail unpredictably under policy guardrails. Use your own photo, a consented subject, or a fictional character brief. This page does not teach celebrity likeness hacks.

How do I export a ChatGPT caricature look to Midjourney or FLUX?

After ChatGPT nails the look, copy the fixed kit plus filled variables into a recreate brief that names the target model. Midjourney v7 wants short phrases with --ar for portrait canvas and --style raw when the kit is photographic-leaning. FLUX wants natural-language materials and light. Keep separate dialect notes per model.

When should I use PromptMake /image for caricature work?

Use it when you hold a reference selfie and need model-ready prompt text for Midjourney, FLUX, DALL·E, SDXL, or Leonardo. Pick Change Style to restyle a portrait upload into caricature language, or Recreate Exactly when the reference already shows the exaggeration you want captured. Open https://promptmake.net/image with a consented still; guest tier allows about three runs per day.

Which ChatGPT model should I use for caricature prompts in 2026?

GPT-5.5 Instant fits quick kit fills and short edits in ChatGPT Image. GPT-5.6 Sol, Terra, or Luna fit longer strip passes when you turn a social paste into fixed slots. Match the text step model to the job: fast iteration in Instant, careful templating in Sol-class chat before you generate the image.