HAIR AI

Role: Co-founder · backend and mobile
Company: Alamet
Platform: iOS · Android
Category: Photo · beauty
Market: Global · multilingual

01Overview

Hair AI answers the question “would this look good on me?” before you go to the hairdresser. Users upload or take a photo of themselves; the app finds the hair region automatically, and they can touch it up with a brush if they want. The chosen style and color are applied to the hair alone, leaving the face and background untouched.

Problem

Most of the time, the only way to really see a new hairstyle is to cut or dye your hair. Off-the-shelf filters distort the face and the photo, don't look realistic, and are limited to a handful of preset styles.

Solution

The hair region is masked automatically, and generative AI redraws only that region (inpainting). The desired hair is described with preset styles, a color palette or free text, and every generation returns multiple variations. Jobs run through a queue on GPU servers, with progress shown live in the app.

02How it works

01
Upload
The photo is uploaded to the cloud, and the user's daily generation quota is checked.
02
Mask
The hair region is detected automatically; the user can refine it with a brush and eraser.
03
Prompt
The chosen style, color or free text is turned into prompts suited to the model.
04
Generation
The job enters the GPU queue; the model redraws only the masked region.
05
Result
Variations come back to the app, where they are rated, saved or shared.

03Highlights

01Automatic hair maskA segmentation service finds the hair region in the uploaded photo. Users can refine the mask with a brush and eraser; the face and background never change.
02Style, color or your own descriptionPreset hairstyles, a wide color palette and free text. Styles and prompts are served from the backend, so adding a new style doesn't require an app update.
03VariationsEvery generation returns multiple results, and any result you don't like can be regenerated with a single tap.
04Subscriptions and daily quotaFree and Pro plans differ in daily generation quota, style and color options, HD results and watermarking. Subscriptions are verified on the server through RevenueCat and App Store notifications.
05Easy sign-inSign in with Apple, Google, Instagram or a code sent by email. Users who start as guests can save their account later.
06Feedback loopEvery result can be rated and commented on; the feedback shows which styles and prompts work well.

04Screens

01WelcomeSign in with Apple, Instagram, Google or email, or continue without an account.

02Pick your photoUpload from your gallery, take one with the camera, or try a sample photo.

03Mark your hairYour hair is detected automatically; fine-tune with the brush and eraser, then choose a style and color.

04GenerationThe request enters the GPU queue, and progress shows up live.

05ResultSwipe through the variations, regenerate if you don't like them, then save or share.

The user journey in five steps, from welcome screen to result.

05Architecture

01Mobile

Sign in with Apple and GoogleIdentity token
Mobile appiOS and Android · Expo
RevenueCatSubscriptions and purchases

02Server

Shared APIJWT · daily quota · app settings
Generation queueFeeds pending jobs to the GPU in order

03Data and AI

MySQLUsers, jobs and subscriptions
AWS S3Photos, masks and results
Hair segmentationAutomatic mask
Serverless GPUStable Diffusion inpainting
GPU serverBackup generation pipeline

The photo reaches the API, the hair mask is extracted, the job is written to the queue, and the app immediately gets a job ID back. The queue hands the job to a GPU and the results are written to S3; the app polls for status and fetches the results directly from S3 through signed URLs. Photart uses the same API and queue.

External service

06Technology