AI hair colour try-on: how to test a shade before you commit

There is a specific kind of regret that only hair colour can produce. A haircut grows out in three to six months. Bad colour either stays for a year or costs another six hours, three sessions, and a four-figure bill to correct. The AI render is the rehearsal that should happen before that booking ever gets made.

This is the colour companion to our pillar AI hairstyle try-on: the 2026 guide — read that first if you want the underpinnings of how the diffusion pipeline works and what identity-preservation means. This piece is about colour specifically: what renders honestly, what doesn't, and how to use the output to brief a colourist so the appointment lands the first time.

#What AI hair colour try-on actually is

Modern AI colour try-on uses the same diffusion architecture as hairstyle try-on, but with the model conditioned on a target shade rather than a target cut. A well-built system does five things in order:

1. Detects your face and hair region — segmentation isolates the hair from the background and the skin.
2. Reads your current hair colour as the underlay. A balayage rendered from black hair should look different than the same balayage rendered from blonde hair; if the system ignores your starting point, the output is fiction.
3. Holds your identity — skin tone, undertone, eye colour, freckles, and the shape of your features are not negotiable.
4. Applies the target shade — including realistic light and shadow within the hair (highlights need to follow the hair's natural fall, not paint flat colour across the strand).
5. Returns an image that looks like a photo of you, lit by the same light source as your selfie. Anything else is a filter, not a try-on.

The differences between the credible tools and the toy filters land almost entirely in steps 2 and 5. The toys skip them and render every shade as if your hair started bleached and the studio light is on.

#The four questions to ask before you book

Colour is more reversible than a haircut in principle — you can recolour — but each recolour is another chemical pass on hair you've already chemically processed. Run these four checks before you book the chair.

#1. Does the shade work with your skin's undertone?

The single biggest reason a hair colour reads "off" is undertone mismatch. Skin undertone falls into three buckets:

• Cool (pink, blue, or red undertone visible through fair skin; veins read blue-purple) — pulls warm shades brassy. Cool blondes (icy, ash, beige), cool brunettes (espresso, mocha), and clear reds work. Golden blondes and copper reds skew orange on you.
• Warm (yellow, golden, peach undertone; veins read green) — pulls cool shades flat or grey. Warm shades (honey, caramel, copper, chestnut, golden brown) read alive on you. Ash blondes and platinum can read sickly.
• Neutral (a mix, veins read both) — most shades work; pick by mood, not by rule.

If you don't know your undertone, the AI render is a faster diagnostic than a Pinterest hour. Run the same shade in a cool version and a warm version of itself on your own face. The one that looks like you is your undertone family.

#2. What is your starting hair colour, honestly?

A colourist's first question, every time, is what level are you starting from? Levels run 1 (jet black) to 10 (lightest blonde), and every shade transition has a chemical cost that scales with how many levels you're moving.

The AI render does not warn you about chemistry. It will happily render you platinum from black in a single image. That image is real as an aspiration and a lie as a Saturday appointment. Bring it as a destination, not a deadline.

#3. What does the render actually look like on your face?

This is where most people skip the test that matters. After you generate the render:

• Cover the hair with your thumb. Does the face still belong to you? Skin tone, eye colour, expression — all yours?
• Look at the temple hair, not the bulk. The temple is where colour misreads first.
• Check the eyebrows. A radical colour change without an eyebrow shift looks wrong in the same way that a wig in the wrong shade looks wrong. The AI won't usually change your brows; your colourist will need to discuss it.

If a render makes you look unfamiliar in your own face, the shade is wrong before the colourist ever opens a tube.

#4. Are you willing to maintain it?

Every colour has a maintenance cost — root touch-ups, gloss, toner, purple shampoo, UV protection, salon-cost — that increases with the distance from your natural colour.

The render does not show you the touched-up version eight weeks in. Mentally render that too before you book.

#Studio prompts that work for common colour decisions

These are the prompts that produce useful renders in the Mademoiselle Studio specifically, but they work as general guidance for any system that takes natural-language colour direction.

Going lighter on dark hair (caramel balayage, brunette base):

"Soft caramel balayage on dark brown base hair, brightness around the face and the mid-lengths, root remains dark, even side light, photographic rendering."

Honey balayage on medium brown:

"Honey-blonde balayage on warm medium-brown base, brightness highest at the cheekbones, root smudged dark, soft natural waves, daylight."

Going cooler — ash brunette:

"Cool ash brunette across this head, slight dimension at the front, no warm undertones, even natural light."

Copper on warm-toned brunette:

"Rich warm copper across this head, slight gloss, brightness at the ends, soft natural waves, daylight."

Platinum from dark hair (the aspirational render):

"Cool platinum blonde across this head, no yellow, slight purple undertone, soft natural waves, studio light. Render as a long-term goal, not a one-session outcome."

The last clause is for you, not the model. Renders of dramatic colour transitions are useful as destinations; treat them as a vision board, not a service order.

Soft grey blend (the most-requested under-the-radar colour right now):

"Cool natural grey blended throughout this head, no yellow, no purple cast, slight dimension, soft waves, even daylight."

Warm red (autumn / copper-red):

"Rich auburn red across this head, slight warmth at the ends, brightness around the face, soft natural waves, golden-hour light."

#Taking the render into the chair

A render is a brief, not an order. Print it (or screenshot it onto a real phone the colourist can hold), and bring two more references:

• A real-photo reference of the same shade on a person whose starting hair colour was similar to yours. The render shows the shade on your base; the real-photo reference shows what the same shade looks like rendered by real light on a real head.
• A real-photo reference of the placement if you're going for balayage or highlights — where do you want the brightness to land? Cheekbones, ends, root, front sections? Photos are clearer than words.

#How starting hair condition affects the result

The AI render does not know whether your hair is virgin (never coloured), previously highlighted, previously bleached, previously single-process coloured, or previously henna'd. Each of those starting points changes what is chemically possible.

• Virgin hair lifts most cleanly. The AI render is closest to one-session reality on virgin hair.
• Previously highlighted hair has banding — old colour zones that lift differently. A balayage render on highlighted hair is honest about the destination but not about the labour to get there.
• Henna'd hair is the colourist's hardest case. Henna binds with the hair shaft and many hair dyes will not lift through it cleanly; some will turn green. Tell your colourist about any henna in your hair history, no matter how long ago. The render will not catch this.
• Bleached hair is fragile. Going darker over heavily bleached hair often grabs darker than expected and fades fast unless a pigment fill is done first. The render shows the steady-state colour; the first two weeks may look much darker.

#Three common mistakes that wreck the render

1. Using a filtered selfie. Beauty filters on iPhone Portrait mode and most social-app cameras pre-warm your skin and oversharpen the hair. A render on a filtered selfie returns a render of someone who isn't quite you. Use the camera app at default, with no portrait blur, in daylight.
2. Yellow indoor light. Warm interior lighting (tungsten bulbs, "warm white" LEDs) pre-tints your hair golden in the input photo. The model then renders the new shade over that warm tint — making cool shades look brassy and warm shades look orange. Stand near a window with daylight, or shoot outside in soft shade.
3. Picking a shade for the render that you wouldn't wear in real life. Run platinum once for fun; then run the three shades that are within reach of your actual hair and your maintenance willingness. Those are the decisions that matter.

#After you have the right render

When you have a render that survives the thumb-over-the-hair test and feels like a sustainable colour, the next reads are:

• How to talk to your stylist — same brief mechanics apply to colourists; the three-photos-and-a-phrase model is the cleanest hand-off.
• AI hairstyle try-on: the 2026 guide — if cut and colour are happening in the same appointment, render the cut on the new colour, not the old.
• Hairstyles by face shape — colour and cut interact. A cool blonde on a sharp square face reads differently than the same colour on an oval — sometimes the cut needs to soften when the colour goes bold.

#Authority and further reading

The colour-level system referenced above is the international standard used by L'Oréal Professionnel, Wella Professionals, and Schwarzkopf Professional — see, for general grounding, the L'Oréal Professionnel education library at lorealprofessionnel.com and the American Academy of Dermatology at aad.org for chemistry-of-hair-dye basics relevant to scalp health.

If a render contradicts what your colourist tells you about your specific hair history, trust your colourist. The render is reading pixels. They are reading hair.