ChatGPT Caricature Prompt: Turn a Selfie Into a Caricature

This guide shows exactly how to write a chatgpt caricature prompt and run it as a selfie‑to‑caricature job on GPT Image 2 inside Omnely. You get a reusable prompt formula, ten‑plus full prompt briefs (job, hobby, and style variants), credit costs down to the cent, and a likeness‑preservation checklist so the character still looks like the person in the photo. The example images below were generated on Omnely with GPT Image 2.

Caricature is a specific craft: keep the identity, exaggerate a few features, pick a medium, and stage a scene. A one‑line viral prompt can work when ChatGPT remembers who you are from previous messages. On Omnely you do not rely on chat memory — you write the identity and the scene into the prompt itself. That is where this post is different from a generic photo library like ChatGPT photo prompts library: everything here is caricature‑first.

How a ChatGPT caricature prompt works on GPT Image 2

A caricature prompt is a short brief with four jobs: lock the likeness, dial the exaggeration, name the medium, and describe the scene. On ChatGPT, people often type a single viral line like "turn me into a Pixar caricature with my job and hobbies" and it works because the chat already knows them. That identity context is the hidden half of the prompt.

ChatGPT's image generation runs on OpenAI's GPT Image model family. Omnely gives you GPT Image 2 from that family, so the same plain‑language caricature‑brief style of prompting carries over. The difference is practical: Omnely has no chat memory, so you must spell out the person — skin tone, hair, glasses, outfit, age range, distinctive features — directly in the prompt. The upside is control. You can swap jobs, hobbies, and styles in seconds without retraining a conversation.

Two ways to run a caricature on GPT Image 2:

• Text to Image — no selfie, you describe the person in words. Good for stylized avatars or when you don't want to upload a face.
• Image to Image — upload a selfie as a reference image (the Studio field reference_image_urls). GPT Image 2 reads the face and uses your prompt to redraw it. Edits cost the same as a fresh generation.

GPT Image 2 on Omnely supports 1 to 16 reference images, 16 aspect ratios, and three output resolutions. There are no quality tiers for this model — the only cost lever is resolution.

Resolution	Credits per image	USD
1K (default)	10	$0.10
2K	15	$0.15
4K	20	$0.20

Image‑to‑Image edits cost the same as text‑to‑image at the same resolution. Aspect ratio does not change the price, but a few ratios clamp output: auto, 4:5, 5:4, 3:1, 1:3, and 9:21 are 1K only; 1:1 tops out at 2K; the rest can go to 4K.

The caricature prompt formula

Every good caricature prompt on GPT Image 2 is built from the same seven parts. Write them in roughly this order and the model will follow.

Part	What it does	Example phrase
Likeness keep rules	Protects identity across the edit	"Keep her warm medium‑brown skin, short natural coiled hair with silver strands, round tortoiseshell glasses"
Exaggeration level	Tells the model how far to push features	"Exaggerate slightly: head about 1.5x body scale, bigger friendly smile, slightly oversized glasses"
Style / medium	Picks the look	"3D Pixar‑like illustration with smooth shading" or "hand‑drawn editorial ink and watercolor on cream paper"
Job or hobby props	Builds the scene around a theme	"standing desk with a glowing laptop, sticky notes, steaming coffee mug"
Pose and framing	Controls composition	"three‑quarter view, chest‑up, slight lean forward"
Background	Keeps focus on the character	"clean cream background with soft warm studio light"
Avoid list	Removes common failures	"no text, no watermark, no extra people, not mean"

A reliable Image‑to‑Image template looks like this:

Using the uploaded reference photo as the exact likeness, create a [STYLE] caricature of this same [person descriptor]. Keep [identity details: skin tone, hair, glasses, outfit, distinctive features]. Exaggerate [subtle/moderate/bold] — specifically [features to push]. Pose them [action and props tied to job or hobby]. [Lighting]. [Background]. Flattering and friendly, not mean. No text, no watermark, no extra people.

A Text‑to‑Image template, no selfie needed:

A [style] caricature portrait of a [age] [gender] with [skin tone], [hair], [glasses or facial features], wearing [outfit]. Exaggerate [features] [subtle/moderate/bold]. They are [job or hobby] — show them [specific action with props]. [Lighting] and [background]. Flattering, cheerful, not caricature-mean. No text, no watermark.

Selfie to caricature on Omnely: step by step

1. Open the Image workspace at Image workspace and pick GPT Image 2 from the model menu.
2. Switch the mode to Image to Image.
3. Upload your selfie as a reference image. GPT Image 2 accepts 1 to 16 references — one clean, well‑lit face shot is usually enough.
4. Paste your caricature prompt. Keep the identity line at the top.
5. Pick an aspect ratio (3:2 and 4:3 are good for desk scenes; 9:16 for phone wallpapers; 16:9 for banners).
6. Pick resolution: 1K = 10 credits ($0.10), 2K = 15 credits ($0.15), 4K = 20 credits ($0.20). Start at 1K to test the prompt, then rerun your favourite at 4K.
7. Generate, review, and iterate. Each edit is a fresh Image‑to‑Image run at the same price.

For the full workspace walkthrough — settings, reference stacks, and troubleshooting — see how to use GPT Image 2.

For a second take on the same selfie, Seedream 5.0 Pro on Omnely also runs image-to-image from up to 10 input images, at 10 credits for 1K or 20 credits for 2K.

For the examples in this post, a synthetic reference selfie was first generated with GPT Image 2 (Text to Image), then that face was used as the reference for every caricature. Here is the exact prompt for the reference face:

Photorealistic casual selfie portrait of a woman in her early thirties with warm medium-brown skin, short natural coiled hair with a few silver strands at the temples, round tortoiseshell glasses, a soft closed-mouth smile, wearing a mustard linen shirt. Soft window daylight from the left, plain pale wall behind, phone-selfie framing chest-up, neutral white balance, visible pores and natural skin texture, no retouching, no beauty filter, no text, no watermark.

Generated with GPT Image 2 on Omnely from the prompt above (4:5, 1K, 10 credits) — synthetic reference face for the caricature edits below.

Settings used: Text to Image, aspect 4:5, 1K, 10 credits ($0.10).

Now the same face as a job caricature, Image to Image with the selfie above as the reference:

Using the uploaded reference photo as the exact likeness, create a playful professional caricature of this same woman. Keep her warm medium-brown skin, short natural coiled hair with silver strands, round tortoiseshell glasses, mustard linen shirt, and identity. Exaggerate slightly: head about 1.5x body scale, bigger friendly smile, slightly oversized glasses. Pose her at a standing desk with a glowing laptop, sticky notes, and a steaming coffee mug — she is a product designer. Soft warm studio light, clean cream background, 3D Pixar-like illustration style with smooth shading, cheerful and flattering, not mean. No text, no watermark, no extra people.

GPT Image 2 Image to Image on Omnely using the selfie above as the reference (4:5, 1K, 10 credits).

Settings used: Image to Image, aspect 4:5, 1K, 10 credits ($0.10).

Job and hobby caricature prompt templates

These are full briefs — not viral one‑liners. Swap the identity block to match your own reference photo.

Nurse on shift

Using the uploaded reference photo as the exact likeness, create a warm 3D illustrated caricature of this same person. Keep hair, skin tone, glasses if present, and face shape. Exaggerate moderately: head about 1.6x body scale, bright smile, slightly larger expressive eyes. Dress them in clean teal scrubs with a stethoscope around the neck and a clipboard in one hand. Hospital corridor softly blurred behind, cool daylight from windows, Pixar-like shading. Friendly and calm, flattering, not stressed. No text, no watermark, no extra people.

Settings: Image to Image, aspect 3:4, 1K, 10 credits ($0.10).

Primary school teacher

Using the uploaded reference photo as the exact likeness, create a cheerful cartoon caricature of this same person as a primary school teacher. Keep identity: skin tone, hair, any glasses, face shape. Exaggerate slightly: slightly oversized head, big friendly smile, animated hands. Pose them beside a chalkboard with simple math and a smiley face drawn on it, holding a stack of colorful notebooks, a red apple on the desk. Warm classroom lighting, soft morning light from a window, flat vector-friendly 3D style, bright but not neon. No text in the final image except doodles on the chalkboard, no watermark.

Settings: Image to Image, aspect 4:3, 2K, 15 credits ($0.15).

Senior backend developer

Using the uploaded reference photo as the exact likeness, create a geeky-friendly caricature of this same person as a senior backend developer. Keep hair, skin tone, glasses, beard or facial hair, outfit colors. Exaggerate moderately: larger glasses, slightly bigger head, intense but kind focus. Three monitors around them showing abstract code and a terminal, mechanical keyboard, a half-empty mug labeled nothing specific, dark room lit by cool screen glow and one warm desk lamp. Semi-realistic stylized 3D caricature, moody but flattering. No readable code text, no logos, no watermark.

Settings: Image to Image, aspect 16:9, 4K, 20 credits ($0.20).

Barista pulling a shot

Using the uploaded reference photo as the exact likeness, create a lively caricature of this same person as a specialty-coffee barista. Keep identity, hair, skin tone, glasses, piercings or tattoos if visible. Exaggerate slightly: head 1.4x scale, confident grin, slightly larger hands focused on the portafilter. Behind an espresso machine with steam curling up, a line of ceramic cups, chalkboard menu softly blurred behind. Warm café light, 3D Pixar-like illustration style, flattering. No readable text, no brand logos, no watermark.

Settings: Image to Image, aspect 3:2, 2K, 15 credits ($0.15).

Home baker (hobby)

Using the uploaded reference photo as likeness, create a hobby-focused caricature of this same woman. Keep her face, glasses, short coiled hair with silver strands. Instead of a desk job scene, show her as an avid home baker: oversized whisk in one hand, flour dust on cheek, colorful mixing bowls and a cooling rack of cookies around her. Slightly exaggerated joyful expression, head a bit large, bright colorful illustration style (not photoreal), flattering and fun. No text, no watermark.

GPT Image 2 Image to Image on Omnely using the selfie as the reference (1:1, 1K, 10 credits).

Settings: Image to Image, aspect 1:1, 1K, 10 credits ($0.10).

Weekend road cyclist (hobby)

Using the uploaded reference photo as the exact likeness, create a dynamic caricature of this same person as a weekend road cyclist. Keep face, hair, glasses swapped for sport sunglasses, skin tone, build. Exaggerate moderately: slightly oversized head inside a bright aero helmet, strong grin, exaggerated forward lean. Pose them riding a colorful road bike on a sunlit country lane, motion lines behind, rolling green hills in the background. 3D illustration style with crisp shading, flattering and energetic. No sponsor logos, no text, no watermark.

Settings: Image to Image, aspect 16:9, 2K, 15 credits ($0.15).

Style variations: Pixar, ink, pencil, watercolor

Changing the medium is the fastest way to get a totally different caricature from the same face. Keep the identity block, swap the style line.

Editorial ink and watercolor

Using the uploaded reference photo, create a hand-drawn editorial ink caricature of this same woman. Keep identity: medium-brown skin tone suggestion in wash, short coiled hair with silver strands, round tortoiseshell glasses, mustard shirt collar. Exaggerate distinctive features gently (smile, glasses, hair volume). Clean black ink linework with soft watercolor washes on warm cream paper, boardwalk caricature artist energy but flattering. Simple background, no text, no watermark.

GPT Image 2 Image to Image on Omnely using the selfie as the reference (4:5, 1K, 10 credits).

Settings: Image to Image, aspect 4:5, 1K, 10 credits ($0.10).

Soft graphite pencil sketch

Using the uploaded reference photo as likeness, create a soft graphite pencil caricature of this same person. Keep identity details intact. Exaggerate subtly: slightly enlarged head, warmer smile, a touch more hair volume. Loose cross-hatched shading, visible pencil texture on off-white paper, no color except faint sepia wash on cheeks and lips. Chest-up framing, soft focus background, flattering and literary. No text, no watermark.

Settings: Image to Image, aspect 3:4, 2K, 15 credits ($0.15).

Felt-tip marker and watercolor mix

Using the uploaded reference photo as likeness, create a bright felt-tip marker caricature of this same person with watercolor fills. Keep face identity, hair, glasses, outfit colors. Exaggerate moderately: head about 1.6x body scale, big smile, slightly oversized hands if a prop is included. Bold black outlines, flat marker colors with light watercolor bleeds, textured watercolor paper background with a soft colored halo behind the figure. Playful, flattering, modern illustration style. No text, no watermark.

Settings: Image to Image, aspect 1:1, 2K, 15 credits ($0.15).

Clay-style stop-motion look

Using the uploaded reference photo as likeness, create a clay-style stop-motion caricature of this same person. Keep identity: skin tone, hair, glasses, outfit. Exaggerate boldly: head nearly 2x body scale, big expressive eyes, oversized hands. Visible clay texture and fingerprint-like surface detail, soft studio key light with a warm fill, shallow depth of field, plain colored seamless backdrop. Charming and flattering. No text, no watermark.

Settings: Image to Image, aspect 4:5, 1K, 10 credits ($0.10).

For non‑caricature portrait, food, and real‑estate style prompts, see the general library at ChatGPT photo prompts library. For a different illustration model's prompt library, see Nano Banana Pro prompt library.

Exaggeration levels and likeness fixes

Caricature lives on a slider between "flattering portrait" and "big‑head cartoon". Say the level in plain words and the model will respect it.

Level	What to write in the prompt	Typical look
Subtle	"Exaggerate subtly: head about 1.3x body scale, slightly bigger smile, keep features close to real proportions"	Stylized portrait, barely a caricature
Moderate	"Exaggerate moderately: head about 1.5–1.6x body scale, larger expressive eyes or glasses, bigger smile, slightly oversized hands if holding a prop"	Classic Pixar‑style caricature
Bold	"Exaggerate boldly: head about 1.8–2x body scale, dramatic smile, strongly oversized distinctive feature, short cartoon limbs"	Boardwalk caricature artist, clay‑style

Keep identity checklist — put every item that matters into the prompt:

• Skin tone (describe warmth and depth, not race labels)
• Age range
• Hair: length, texture, color, parting, any silver or dyed sections
• Glasses: shape, frame material and color
• Facial hair, moles, freckles, piercings, tattoos
• Outfit color and material (helps with recognisability)
• Build (slim, athletic, broad) if it should carry over

Common problems and fixes:

Problem	Fix line to add
Face drifts, doesn't look like the person	"Match the uploaded reference as the exact likeness; preserve face shape, nose, and smile exactly; do not generalize features"
Output is too realistic, not a caricature	Add explicit style: "3D Pixar‑like illustration with smooth shading" or "hand‑drawn ink caricature on paper" and name an exaggeration level
Exaggeration feels mean	"Flattering and friendly, not mean, no unkind exaggeration, keep proportions playful"
Wrong or missing job props	List props individually: "glowing laptop, two sticky notes, one steaming coffee mug" instead of "desk stuff"
Yellow or orange skin tint	"Neutral white balance, keep skin tone true to reference, no orange or yellow colour cast"
Glasses disappear or change shape	"Keep the exact round tortoiseshell glasses from the reference, do not replace or remove them"
Extra people appear	"Single subject only, no extra people, no background characters"

If a run is close but one detail is wrong, do not restart — rerun Image to Image with the first output plus the original selfie as two reference images, and tighten only the broken line. Each new run at 1K is 10 credits ($0.10).

FAQ

Does this work without ChatGPT chat history?

Yes. Omnely has no chat memory tied to a face. You write identity details directly into every prompt. That is why the templates above always start with a keep‑identity line.

What does a caricature cost in credits?

GPT Image 2 charges 10 credits ($0.10) at 1K, 15 credits ($0.15) at 2K, and 20 credits ($0.20) at 4K. Image‑to‑Image edits are the same price as fresh generations at the same resolution. Aspect ratio does not change the price. A typical test‑then‑polish flow — three 1K drafts and one 4K final — runs 50 credits ($0.50).

How is this different from the ChatGPT photo prompts post?

ChatGPT photo prompts library is a general photo library — portraits, food, real‑estate, product shots. This post is caricature‑only: identity preservation, exaggeration levels, and stylized media. If you want photoreal, use that one. If you want a stylized character of a real person, use this one.

Try the prompts on GPT Image 2

Pick one job and one hobby template, paste in your own identity block, and run them Image to Image on GPT Image 2 on Omnely. Start at 1K for 10 credits ($0.10) per test, then rerun your favourite at 4K for 20 credits ($0.20). The whole formula — likeness keep, exaggeration level, medium, props, pose, background, avoid list — is reusable across every caricature you will ever brief. Head to GPT Image 2 on Omnely to generate the first one.