Prompt library

Pastel Portrait Photo Prompt

Use an identity-first image-editing prompt to turn a casual photo into a powder-blue, gouache-and-pastel editorial portrait while keeping the person recognizable.

By Grow AI Skills Editorial TeamPublished Sep 8, 2026Verified Sep 8, 2026

Release card Behavior Tested

Version: 1.0
Origin: Adapted
Works with: ChatGPT Images with a permitted portrait upload, GPT Image 2 image editing, Codex built-in image generation

Hand-painted pastel editorial portrait of a fictional woman with a small cheek mole, dark wavy hair, closed-mouth smile, warm ivory sweatshirt, and powder-blue background — Final copy-prompt run. The generic reusable prompt retained the visible cheek mole, facial relationships, closed-mouth half-smile, head angle, hairstyle silhouette, flyaways, and clothing type while producing the intended chalky powder-blue illustration. Credit: Original synthetic image edit by Grow AI Skills. Open full-size image.

Best for

Casual adult selfies and phone portraits with a readable face, expression, hairstyle, and upper-body pose
Warm editorial avatars, personal profile illustrations, greeting-card portraits, and private keepsakes
Photos whose background is unhelpful but whose person is still clear enough to describe and review
One-person portraits where identity preservation matters more than exact fabric, background, or photographic texture

Do not rely on it for

Generative editing can change face shape, eye spacing, gaze, age cues, hair, expression, pose, clothing, crop, or visible marks even when the prompt says to preserve them
The exact reusable prompt was tested once on one synthetic adult woman; it was not benchmarked across ages, skin tones, genders, group portraits, occlusion, or different image models
A recognizable editorial result is a human judgment and is not biometric proof of identity
A source that is very blurred, blocked, tiny, or heavily compressed does not contain enough visible information for this prompt to recover missing details truthfully
The prompt does not grant consent or rights to upload, edit, publish, advertise with, or sell a portrait of another person
The powder-blue palette and warm skin colors are a deliberate art direction, not a color-accurate reproduction of the source

Paste-ready prompt

Copy, then replace the variables.

Prompt v1.0

STYLE TRANSFORM ONLY
Use the uploaded photograph as Image 1, the only edit target and identity reference. Turn it into a warm contemporary hand-painted pastel editorial portrait. Do not create a different person and do not perform a beauty makeover.

PRESERVE
Keep the subject recognizably the same: facial proportions and distinguishing characteristics; face shape; eye, eyebrow, nose, and mouth relationships; visible moles or freckles; age and gender presentation; head angle; gaze; exact emotional expression and open-or-closed mouth state; hairstyle, parting, silhouette, direction, and important loose strands; pose; shoulder orientation; visible clothing type; and the source portrait's overall orientation. Do not beautify, slim, reshape, symmetrize, rejuvenate, or replace the face.

ALLOWED CHANGES
Change the artistic medium and replace the photographic environment. A modest crop correction is allowed only when needed to make the head and upper torso dominant. Do not mirror or re-pose the subject.

BACKGROUND
Replace the source environment with one uninterrupted matte powder-blue painted background. Allow only faint natural pigment variation. Add no furniture, architecture, scenery, pattern, gradient, text, symbol, frame, collage, logo, signature, or watermark.

ILLUSTRATION STYLE
Make the result unmistakably illustrated rather than a filtered photo or realistic painting. Combine opaque matte digital gouache shapes, chalky dry-pastel grain, sparse soft colored-pencil marks, irregular dry-brush boundaries, slightly imperfect organic linework, gentle asymmetry, and restrained pigment variation. Reduce photographic detail substantially. Let the face, hair, clothing, eyes, nose, and lips read as deliberately simplified editorial shapes. Keep the result warm, playful, clean, and handmade.

FACE
Retain identity and expression while simplifying anatomy. Enlarge the eyes only very slightly for clarity, never anime-like. Use warm off-white sclera, simple very dark warm-brown pupils, one tiny natural highlight, irregular upper-lash strokes, minimal iris detail, and faint peach-coral eyelid accents. Paint eyebrows as textured dark-brown brush strokes. Describe the nose mainly with one broad warm peach-orange shape, a subtle bridge, and minimal nostril marks. Preserve the mouth's exact expression and open-or-closed state; paint the lips as a simple irregular matte coral shape. Use luminous matte light ivory-peach or pale-apricot skin, broad minimal shading, soft salmon cheek color, and restrained warmth around the nose and eyelids. Remove pores, shine, photographic microtexture, and hard realistic shadows without flattening the face into vector art.

HAIR AND CLOTHING
Build the hairstyle from a few broad opaque warm chocolate, espresso, and deep reddish-brown masses. Add only selected scratchy dry-brush and colored-pencil strands, especially important flyaways. Keep the original hairstyle silhouette and parting. Make the edges irregular and matte, never glossy or strand-by-strand realistic. Preserve the clothing's basic type and shape; simplify it strongly, use warm ivory instead of sterile white when appropriate, and show folds with only a few soft gray-beige marks. Keep clothing lower-detail than the face and hair.

PALETTE
Use powder sky blue, light ivory-peach, pale apricot, soft salmon pink, warm peach-orange, muted coral red, warm chocolate brown, espresso brown, very dark warm brown, and creamy off-white. Keep the colors bright, matte, warm, harmonious, and moderate in contrast.

PRIORITY
1. recognizable identity
2. original expression and gaze
3. hairstyle silhouette and loose-hair gesture
4. head angle, pose, and shoulder orientation
5. clearly simplified editorial shapes
6. chalky handmade texture
7. clean powder-blue background

AVOID
Photorealism; semi-realistic AI glamour painting; generic attractive faces; realistic skin or hair rendering; 3D or Pixar-like rendering; anime; manga; cel shading; vector geometry; comic-book ink outlines; giant eyes; glossy skin; plastic gradients; hyper-detailed pores; individual realistic hair strands everywhere; heavy cinematic light; hard shadows; high contrast; neon colors; dark or complex backgrounds; perfect bilateral symmetry; face slimming; eye-color changes; makeup makeovers; extra people or objects.

Return one finished portrait, not a comparison collage.

Expected output: One portrait-oriented, hand-painted pastel editorial illustration of the uploaded person on a clean powder-blue background.

An unflattering photo is not a broken photo. If the face, expression, hairstyle, and pose are still readable, a careful style transformation can replace harsh lighting and a distracting room with a warm pastel illustration. The useful trick is not “make me prettier.” It is a strict contract: preserve the person, permit only the medium and background to change, and reject any result that wins style points by inventing a different face.

The copy box above contains the exact generic prompt used for the featured result. We ran it from the original synthetic portrait rather than from a previous edit. At manual review, the small cheek mole, eye and brow relationships, closed-mouth half-smile, head tilt, gaze, hair part, wavy silhouette, loose strands, shoulder direction, and crew-neck sweatshirt remained recognizable. The result also moved decisively away from a filtered photo and toward matte gouache, dry pastel, and colored-pencil marks.

Use the prompt as an edit, not a generation request

Choose a portrait you own or have permission to edit. Avoid a photo containing private screens, addresses, badges, documents, or people who did not agree to the upload.
Upload the portrait and make it the only image reference. If the interface shows several images, identify the target as Image 1.
Paste the prompt without asking for a named artist's style. The medium, texture, palette, and shape rules already provide a specific visual language.
Open the original and result side by side at full size. Review identity first and illustration strength second.
If a defining feature moves, restart from the original. Do not keep editing a damaged derivative.

Record a five-line identity lock first

The master prompt can inspect the uploaded image, but a short anchor card makes your own review more disciplined. Write only details visible in the source. Do not guess eye color from blur, reconstruct a hidden ear, or invent a hairstyle behind an occlusion.

FACE RELATIONSHIPS: rounded oval face; almond eyes; natural thick eyebrows
VISIBLE MARKS: small mole below the left cheekbone
EXPRESSION AND GAZE: gentle closed-mouth half-smile; looking at camera
HAIR: off-center part; shoulder-length dark waves; visible flyaways
POSE AND CLOTHING: head tilted right; angled shoulders; white crew-neck top

Use this card as a checklist, not as a second beauty prompt. It describes what the edit must preserve. In our synthetic source, the mole and closed-mouth expression were particularly useful because they are easy to inspect and easy for a generic makeover to erase.

Synthetic casual smartphone portrait of a fictional adult woman with dark wavy hair, a cheek mole, warm-white sweatshirt, mixed apartment light, and a hallway background — Synthetic edit target. Its review anchors were recorded before any style transfer: cheek mole, eye and brow relationships, closed-mouth half-smile, head tilt, gaze, hair part and silhouette, loose strands, shoulder direction, and crew-neck sweatshirt. Credit: Original synthetic image by Grow AI Skills. Open full-size image.

Why the prompt is divided into contracts

Section	Decision it controls	Failure to reject
Preserve	Locks identity, expression, hair, pose, clothing type, and orientation	A smoother, younger, slimmer, more symmetrical, or simply different person
Allowed changes	Limits freedom to medium, background, and a modest crop correction	A new pose, mirrored face, changed mouth, fashion restyle, or invented prop
Illustration style	Specifies matte shapes, chalky grain, dry edges, and sparse pencil marks	A photographic filter, airbrushed glamour painting, 3D render, or vector avatar
Face	Simplifies micro-detail while retaining facial relationships	Giant eyes, pinched nose, cosmetic makeover, erased mark, or plastic skin
Priority	Resolves conflicts in favor of identity before decoration	Beautiful texture attached to the wrong person
Avoid	Names common visual attractors that compete with the target	Anime, cinematic shadows, neon color, glossy strands, or generic doll features

This division matters because “pastel portrait” describes a surface, not the edit boundary. Without a preserve contract, a model can satisfy the style request by drawing a new attractive character. Without allowed changes, “preserve the person” can be interpreted loosely while the crop, clothes, pose, or expression drifts.

What changed across three edit attempts

The first retained edit used detailed identity rules, but its broad facial modeling and many realistic hair strands remained close to a polished portrait painting. It was attractive and recognizable, yet it did not fully satisfy the more graphic, handmade editorial target.

Conservative painterly edit of the synthetic woman with recognizable features, warm brown hair, ivory sweatshirt, and a pale-blue textured background — First retained edit. It preserves the person well, but realistic facial modeling and detailed hair keep it close to a polished portrait painting rather than the stronger simplified editorial target. Credit: Original synthetic image edit by Grow AI Skills. Open full-size image.

The next scene-specific iteration made one creative correction: it demanded flatter color shapes, much less nose modeling, and only selected hair strokes. That moved the result farther from photorealism without changing the source reference. We deliberately restarted from the original synthetic image rather than stacking another generative edit on the first result.

More strongly simplified pastel-and-gouache portrait of the same fictional woman on powder blue, with matte peach skin and rough warm-brown hair marks — Scene-specific iteration. Stronger instructions for simplified shapes, minimal nose modeling, and selective hair strokes moved the result farther from photorealism while retaining the main identity anchors. Credit: Original synthetic image edit by Grow AI Skills. Open full-size image.

The final run removed all person-specific descriptions from the instruction and used the exact generic copy prompt published here. It still retained the source's review anchors. That supports the reusable structure for this one scenario, but it does not prove identical behavior on another face, model, crop, or service.

Run	Identity review	Medium review	Editorial decision
Conservative	Main anchors retained	Too close to realistic portrait painting	Keep as a useful boundary example
Stronger scene-specific	Main anchors retained	Clearer simplified shapes and dry marks	Accept as an iteration example
Exact copy prompt	Main anchors retained	Chalky, simplified, powder-blue editorial result	Feature and publish

Review identity before deciding that it looks good

A result can be charming and still fail. Check these items in order. Stop at the first identity failure; texture polish cannot repair a changed person.

Face shape: does the jaw, cheek width, and forehead relationship still read like the source?
Feature spacing: compare eyes, brows, nose, and mouth as a group rather than reviewing isolated details.
Visible marks: verify meaningful moles, freckles, scars, or other distinguishing marks that were actually visible.
Expression: confirm the emotional character and exact open-or-closed mouth state.
Gaze and head angle: reject a changed eye direction, mirrored tilt, or corrected posture.
Hair structure: compare the part, outer silhouette, length, direction, and a few important loose strands.
Pose and clothing: keep shoulder orientation, crop logic, neckline, and basic garment type.
Illustration strength: only after identity passes, look for matte shapes, restrained detail, rough edges, and selective pencil texture.
Background discipline: confirm a clean powder-blue field with no invented room, text, watermark, or decoration.
Final-size readability: inspect the actual avatar, card, or print size; a mole or hair gesture visible at 100% may disappear in a small crop.

Repair one failure without reopening the whole image

When the result is nearly correct, name one defect and repeat the protected anchors. A targeted correction is easier to judge than another long creative rewrite.

Texture-only repair

Change only the rendering style. Reduce realistic facial modeling and individual
hair strands. Use flatter matte gouache shapes, chalky dry-pastel grain, sparse
colored-pencil marks, and rough handmade edges. Keep the same face, expression,
gaze, visible marks, head angle, hairstyle silhouette, pose, crop, clothing,
powder-blue background, and palette unchanged.

Background-only repair

Change only the background to one uninterrupted matte powder-blue painted field.
Remove every room detail, object, pattern, gradient, letter, symbol, frame, logo,
and watermark. Keep the subject, face, expression, hair, pose, clothing, crop,
colors, and existing illustration marks unchanged.

Restart when the face or expression has drifted

Do not repair identity by repeatedly editing the wrong face. Return to the original photograph, attach it again, and name the failed anchor. The original remains the authority; a convincing derivative is not a new source of truth.

Identity restart

Start again from Image 1, the original photograph. The previous result changed
[FAILED IDENTITY ANCHOR]. Preserve [VISIBLE SOURCE DESCRIPTION] exactly, along
with the original face relationships, expression, gaze, head angle, hairstyle,
pose, and clothing type. Apply the same pastel editorial medium and powder-blue
background, but do not reuse the previous derivative as a reference.

Know when the source is too weak

This workflow can reinterpret a poorly lit or visually plain portrait. It cannot truthfully recover a face hidden by motion blur, tiny resolution, an obstruction, or extreme compression. When a defining detail is absent, the model must invent it. Use another source photo, keep the output intentionally loose, or work with an illustrator who can ask for additional references.

Group portraits are also a separate problem. Several faces increase the chance of swaps, merges, changed gaze, and inconsistent simplification. Start with one adult subject. Treat children, family images, client work, memorial portraits, advertising, or public posting as consent-sensitive uses and confirm the necessary rights before upload and publication.

Limits of this test

All four images on this page are synthetic. The source portrait depicts a fictional adult, and the three derivatives were generated from it with the built-in image tool. The tool did not expose the underlying image-model identifier. We did not run the prompt in the ChatGPT web editor, the GPT Image API, or another provider, and we did not test a second person.

One editor performed a visual comparison. No face-recognition or biometric system was used, and none should be inferred. The result supports a practical preservation checklist and an exact copy-prompt example; it does not establish deterministic repetition or universal identity fidelity. The public test record provides the original and output hashes, exact reusable prompt, call lineage, review anchors, and explicit evidence boundary.

For a broader method that designs a visual language without imitating a named artist, use the Hand-Drawn Style Director Skill. If your actual goal is a natural-looking photograph rather than an illustration, use the separate Unretouched Smartphone Portrait Prompt.