Per generation, at a rate that depends on the model — an edit from a reference image usually costs a little more than a plain generation on the same model. Each model page carries its own rate.

Change a hairstyle with AI and stay recognisable.

Q: Will it still look like me?

On the four models we tried with one portrait, yes — a mole, freckles and a distinctive brow shape survived every edit, including a shave to a buzz cut. Name those features in the instruction and check them in the result: they are the things worth looking at, more than whether the haircut is pretty.

Q: Which model should I use?

Any of the editing models will recut a photograph. The one that mattered in our test was whether a model gives the image back at the size you sent it: GPT Image 2 did, and the others returned a larger canvas — one of them as a JPEG from a PNG. If you are building a set to flick through, that consistency is worth more than a marginally nicer haircut.

Q: How do I try several styles without the face drifting?

Reference the original portrait every time rather than editing the previous result. Each edit then starts from the same face, so the tenth style is as close to you as the first. Editing an edit compounds whatever the last pass changed.

Q: Can it change colour as well as cut?

Yes, and they are separate asks — name the colour and the length in the same sentence if you want both. The copper version on this page is the same portrait with the colour and the length named together.

Q: What kind of photo works best?

A straight-on, evenly lit head-and-shoulders shot, sharp on the eyes, with the whole hairline visible. Harsh side light, heavy shadow across the face and a cropped forehead all leave the model guessing at the thing you want kept.

Q: Can I use this on someone else’s photo?

Only with their agreement. Editing a recognisable person’s appearance is exactly the case where consent matters, and a salon or app putting this in front of clients should be getting it explicitly rather than assuming it.

Q: Do I need to write code to use this?

No — Hedra Studio runs the same models on the same engine with nothing to build. Drop in the photo, name the cut, and compare it against the original.

A try-on is only useful if every version is unmistakably you. Name the cut, name what must not change, and get one face back with a different haircut on it — not a stranger who suits the style.

Start building now Try a hairstyle — no code

Portrait in

A jaw-length blunt bob with a straight fringe, asked for in one sentence. Switch back and check the mole, the freckles and the eyebrows — those are the things a good edit leaves alone.

Editing regenerates the whole frame rather than masking the hair, so skin texture can soften and the background can shift. Name what must not change, and compare against the original before you keep it.

A woman with a blunt black bob and full straight bangs, in a gray T-shirt against a plain wall — GPT Image 2 · 768 × 1024 · canvas kept

A try-on is only useful if it is still you.

A woman with a rounded dark bob and wispy bangs, in a gray T-shirt against a plain wall — GPT IMAGE 2 / KONTEXT MAX / NANO BANANA PRO

Same sentence, three models, three haircuts.

Every one of these is the same portrait and the same instruction, and every one held her face — the mole, the freckles, the eyebrows. What differs is the haircut each model thinks that sentence describes, and something less obvious: only one of the three returned the image at the size it was given. The others handed back a bigger canvas, and one model elsewhere in the catalogue quietly returned a JPEG from a PNG. For a single edit that is trivia. For a set you flick between, the frame jumps and there is nothing left to compare.

A woman with a buzz cut, in a gray T-shirt against a plain wall — GPT IMAGE 2 · BUZZ / COPPER / LOW BUN · ONE REFERENCE

A woman with long wavy auburn hair, in a gray T-shirt against a plain wall — GPT IMAGE 2 · BUZZ / COPPER / LOW BUN · ONE REFERENCE

One face, a whole set.

This is what the page is actually for: not one edit, but a set built from a single portrait so the only thing changing is the hair. A buzz cut, a colour, a pulled-back bun — three answers to three different questions, and the same woman in all of them, freckles and mole intact. Because each one references the original rather than the previous edit, nothing drifts as the set grows.

A woman with long wavy black hair, in a gray T-shirt against a plain wall — Give her a nice modern haircut that suits her.ORIGINAL / UNNAMED / NAMED

A woman with shoulder-length layered dark hair, in a gray T-shirt against a plain wall — Give her a nice modern haircut that suits her.ORIGINAL / UNNAMED / NAMED

Ask vaguely and it gives you back your own hair.

Her own hair, then “give her a nice modern haircut that suits her”, then a named cut. The middle one is not a failure and it is not ugly — it is a competent layered lob. It is also, within a few centimetres, the hair she walked in with. Asked to choose for you, the model picks the safe answer, which is the one answer a try-on cannot use. Name the length, the fringe and the parting and you get the cut you were actually curious about.

Every model that will recut a photograph.

OpenAIGPT Image 2GPT Image 2.5 FlareGPT Image 2.5 Sunburst

Black Forest LabsFlux Kontext MaxFlux Kontext ProFlux.2 [max]Flux.2 [pro]

GoogleNano Banana ProNano Banana 2Nano Banana

ByteDanceSeedream 5.0 ProSeedream 5.0 LiteSeedream 4.5

See all models →

Built for visual inference.

Serving video is a different problem from serving text — a single request can saturate a GPU, and none of the tricks that made language models cheap apply. Hedra's engine was built for exactly that workload.

Whichever model you choose — ours or anyone else's — it runs on the same infrastructure, behind one key. Which is what makes a real try-on affordable: send one portrait and six instructions in parallel, put two models side by side on the same face, and keep whichever held it best. Switching between them is a string, not a migration.

Generate with API or Agent

Build with the API

One key, every editing model. Send one portrait and a list of cuts, and get a set back.

Start building now

Try one now.

No code — drop in a photo, name the cut, and compare it against the original before you keep it.

Create with Agent

FAQs

What should I actually write?

The cut, in the words you would use to a hairdresser: the length, the fringe, the parting, the colour. Then the things that must not change — the face, the expression, the background. Asking for “something that suits me” gets you a safe answer close to the hair you already have, which is the one result a try-on cannot use.

Will it still look like me?

Which model should I use?

How do I try several styles without the face drifting?

Can it change colour as well as cut?

What kind of photo works best?

Can I use this on someone else’s photo?

How is it billed?

Do I need to write code to use this?

More you can do with Hedra

Change anything else in the frameAI PHOTO EDITOR →

Keep one face across a whole setAI CHARACTER GENERATOR →

hedra keys create
→ sk-hedra-••••••••

Put a try-on inside your own appDEVELOPER PLATFORM →