How to construct hair prompts that actually work, styling hair with AI without losing your face.
Want to see yourself with different hair?
RunwayML is surprisingly adept at faces and hair, but only if you don't leave it to guesswork. That's why we always start with one fixed point: the standard Zappa image. A clear portrait of the face, with the hair as it currently is. We upload that image first. Not to change, but to establish who you are.
Many AI tools quickly steer faces towards a kind of average beauty. Runway does this less, especially when you provide it with a good starting image. This way, the identity remains intact, creating space to play with hair: a bit shorter, a bit looser, wet, beachy, or very sleek — without the face shifting.
In this session, we show you how we build hair prompts. Not with grand words, but with attention to what truly matters: length, parting, tension, texture, shine, gravity. Equally important: what the hair should NOT do. That's often where the control lies.
By consistently starting from the same base image and describing the hair step by step technically, calmness emerges in the result. No surprises. No 'different person'. Just another version of the same face.
So, do you want to know how you look with a sleek bob, sun-dried hair, or a more editorial look?
The difference lies not in the tool, but in how you observe, how you describe, and where you begin.
*ChatGPT can be trained very well in hair prompts with this foundation. Provide GPT with the prompt examples below and then a photo of the hair you want to see on YOUR head, and use those prompts for Runway.

Subtle hairstyle adjustment only.
Keep face identity, facial structure, skin texture, expression, lighting, camera framing and background exactly unchanged.
Change only the hairstyle:
– hair slightly shorter, ending around collarbone
– center part (middle part) remains clean and symmetrical
– introduce soft natural waves / loose curls, low volume
– realistic curl pattern, imperfect and organic, not styled
– hair color unchanged
No changes to face, body, clothing, pose or environment.
No beautification, no makeup changes.
Photorealistic result.

High-fashion studio portrait.
ZAPPA identity locked to reference image.
Hair long, layered, and softly textured, falling below the shoulders.
Natural center part with slight irregularity, not razor-straight.
Hair color deep dark blonde to light brown, muted and natural, with subtle tonal variation and slightly darker roots.
Texture soft and airy with loose, natural waves, uneven and relaxed — not styled, not curled.
Front sections fall forward, framing the face with gentle asymmetry.
Finish matte-to-satin, no gloss, no wet look, no beauty polish.
Hair follows gravity and natural movement, not salon styling.
Soft, controlled studio lighting shaping face and hair texture.
Dark, near-black studio background maintained.
No changes to face, skull shape, skin tone, proportions, or identity.
How to Construct Effective Hair Prompts for RunwayML
What many people do wrong:
they start with the hairstyle.
What actually works:
you build a prompt in layers, from stable → variable.
Think not in terms of “prettier”, but in physical behaviour.
Step 1 — Always Lock the Identity First
This is not optional.
ZAPPA identity locked to reference image.
Why:
Faces are fragile. If you don't explicitly lock this, Runway compensates as soon as you change the hair. Then you get “a different person with the same hairstyle”.
👉 Rule: no hair without identity lock.

Subtle hairstyle transformation only.
Keep face identity, facial structure, skin texture, expression, lighting, camera framing and background exactly unchanged.
Change only the hairstyle:
– hair becomes longer, falling past the shoulders
– center part (middle part), clean and symmetrical
– hair texture natural and realistic, soft movement, no volume exaggeration
– hair color unchanged
No changes to face, body proportions, clothing, pose, or environment.
No beautification, no stylization, no makeup changes.
Photorealistic result.

High-fashion studio portrait.
ZAPPA identity locked to reference image.
Hair tightly slicked straight back, clean hairline, no loose strands, wet-look finish.
Neutral expression, direct gaze.
Controlled directional studio lighting, deep shadows, black background.
No changes to face, skull, skin tone, or proportions.
Step 2 — Determine Hair Length in Absolute Terms
Not: short, medium, bob.
Yes: measurable and anatomical.
Examples:
just above shoulder length
at jawline length
between jawline and cheekbone
Why:
AI understands the body better than fashion terms. You link hair to bone structure, not trends.
Step 3 — Parting = Geometry, Not Style
Never just say “middle part”.
Describe:
position (center / off-center)
precision (clean / broken / irregular)
behaviour (tight / natural / disrupted)
Example:
clean middle part, razor-straight
irregular off-center part, broken and uneven
Why:
The parting determines facial symmetry.
This is crucial for recognition.

High-fashion studio portrait.
ZAPPA identity locked to reference image.
Hair worn loose and natural, shoulder-length, soft straight texture with light volume, subtle side part, slightly tousled ends.
No slicking, no wet look.
Neutral expression, direct gaze.
Soft controlled studio lighting.
Dark, minimal studio background maintained.
No changes to face, skull, skin tone, or proportions.

High-fashion studio portrait.
ZAPPA identity locked to reference image.
Hair worn loose and natural, shoulder-length blunt cut, clean middle part, smooth straight texture with light volume, ends slightly tucked under.
No slicking, no wet look.
Neutral expression, controlled gaze.
Soft, directional studio lighting.
Dark, minimal studio background maintained.
No changes to face, skull, skin tone, or proportions.
Step 4 — Describe Hair Texture as Material
Think like a stylist + photographer.
Use words like:
matte / satin / localized sheen
strand separation
clumping / fragmentation
dry–salty / sweat-touched
surface sheen vs. internal shine
Avoid:
pretty
glossy
styled
beautiful
Why:
Runway responds to material properties, not aesthetic judgments.
Step 5 — Explicitly State What is NOT Allowed
This is essential.
Examples:
no volume lift
no waves
no flyaways
no grease
no symmetry
Why:
AI fills in gaps on its own. “Not mentioned” = “free play”.

High-fashion studio portrait.
ZAPPA identity locked to reference image.
Hair cut just above shoulder length, clean blunt line.
Precise middle part.
Both sides styled tightly behind the ears, ears fully visible.
Hair texture smooth and controlled with subtle wet-look definition — light surface sheen, separated strands, no dripping, no grease.
Roots slightly darker and denser, lengths lighter with fine strand separation.
Ends straight and compact, no flicking outward.
No volume lift, no waves, no flyaways.
Hair follows gravity and head shape exactly.
Neutral expression, calm direct gaze.
Directional studio lighting emphasizing hair texture and facial structure.
Dark, minimal studio background maintained.
No changes to face, skull shape, skin tone, proportions, or identity.

High-fashion studio portrait.
ZAPPA identity locked to reference image.
Hair worn loose and natural, shoulder-length blunt cut, clean middle part, both sides neatly tucked behind the ears, smooth straight texture with light volume, ends softly tucked under.
No slicking, no wet look.
Neutral expression, controlled gaze.
Soft, directional studio lighting.
Dark, minimal studio background maintained.
No changes to face, skull, skin tone, or proportions.
Step 6 — Let Hair Respond to Natural Forces
This is where prompts suddenly become realistic.
Good prompts say:
hair follows gravity and heat
shaped by wind, not styling tools
ends bent by movement
Poor prompts say:
perfectly styled
salon finish
Why:
Runway is extremely good at simulating physics — use that.
Step 7 — Light and Background After the Hair
Always this order:
identity
hair shape
hair texture
behaviour
light
background
Not the other way around.
Why:
Light influences how hair is perceived (matte vs gloss, volume vs flat).

High-fashion studio editorial portrait.
ZAPPA identity locked to reference image.
Hair short, textured bob, length between jawline and cheekbone.
Deep off-center part, irregular and broken.
Hair appears sun-dried and sweat-touched — matte base with localized sheen.
Uneven strand distribution:
– one side falls forward across the forehead and cheek, partially obscuring one eye
– the other side is pushed back and slightly lifted at the root
Visible separation in clumps, natural disorder, no symmetry.
Texture is dry–salty, not styled smooth, not glossy.
Ends sharp, fragmented, slightly bent by movement.
Hair follows gravity and heat, not styling tools.
Lighting emphasizes skin warmth and hair texture with directional highlights.
Dark, industrial background retained.
No changes to face, skull shape, skin tone, proportions, or identity.

High-fashion studio editorial portrait.
ZAPPA identity locked to reference image.
Hair cut at jawline length, blunt but not symmetrical.
Clean middle part with slight natural deviation, not razor-straight.
Hair color natural blonde with subtle tonal variation.
Hair behavior is editorial, not beauty:
– one side falls straighter and heavier
– the opposite side shows a soft bend and uneven tension
– no mirrored shape, no perfect balance
Texture matte-to-satin, no glossy finish, no commercial polish.
Hair sits close to the head, no volume shaping, no salon blow-dry effect.
Ends slightly irregular, following gravity rather than styling rules.
No symmetry. No beauty catalog hair. No perfect bob.
Soft controlled studio lighting.
Dark, minimal background.
No changes to face, skull shape, skin tone, proportions, or identity.
The Core Summary
Good hair prompts are not beauty text.
They are technical descriptions of behaviour.
Zappa therefore does not work with:
“beautiful hair”
but with:
“how hair behaves under light, gravity, and tension — without changing the face.”
And that is why RunwayML can be so good at faces and hair,
provided you tell it how it really works.
Regards, Peter & Zappa.