Mirra镜衣
v0.19.0See the changelog →

OPEN SOURCE · AI TRY-ON

New face.New outfit.New pose.

Mirra is an open-source AI try-on app. Pick a model face, choose an outfit, set a pose, add a background if you like, and generate your look in one tap. Requests only travel between your device and the AI service you set up.

  • Free, no account
  • Runs with your own server
  • Keys stay on your device
01 FACE
Try another face →

Model faces · CC0

HOW IT WORKS

Four steps to your look

01

Pick a face

Choose one from the CC0-licensed AI model library, filtered by short, medium, shoulder-length or long hair.

SpringSummerAutumnWinterPartyMy clothes

02

Choose an outfit

Browse complete looks by season, party and swimwear, keep the ones you like with ♡, or add photos of your own clothes laid flat or on a hanger.

FrontThree-quarterSideFrom aboveFrom below

03

Set a pose

Pick a pose from the mannequin line-art library and choose front, three-quarter, side, high or low angle, or make one sheet of the pose from every angle.

04

Choose a background

Optional. By default the style sets the scene: studio, street, white backdrop, film or anime. Or pick one of the built-in indoor, city and nature backgrounds. Then generate, or let Surprise me put a look together for you.

WHY MIRRA

Your datastays yours

Mirra is designed never to touch your data: no account, no servers of ours, and no one’s photos.

Open source & free

All code is public under MIT. No fees, no sign-up.

Bring your own AI

OpenAI, Google Gemini, Volcengine Ark, xAI, OpenRouter, EveryAPI, Alibaba Cloud Model Studio and Azure OpenAI are preset: choose one and enter its key. Or connect any OpenAI-compatible service, or run stable-diffusion.cpp on your own machine at no per-call cost.

On your device only

API keys stay in the system keychain; photos of your clothes and your looks stay on your device. Requests go straight to the service you set up.

One experience, everywhere

Phone, desktop and browser share the same flow, including wardrobe, lookbook and AI settings, in eleven languages. The web version keeps its key and looks only as long as the page.

QUICK START

Up and running in minutes

Mirra ships without a model. Connect any cloud API, or a service you run yourself. Pick a path:

1

Download & install

Pick your platform below, then download and install Mirra.

2

Open AI service

Open Me on a phone, or the settings button at the top right on a desktop and then AI service, and choose a provider.

3

Enter a key & test

Enter your API key (Get an API key opens the provider’s page), choose a model from the provider’s list, press Test connection, and you are ready to generate. The address and size are preset; change them under Advanced if you need to.

Supported AI services:OpenAIGoogle GeminiVolcengine ArkxAIOpenRouterEveryAPIAlibaba Cloud Model StudioAzure OpenAIstable-diffusion.cppOpenAI-compatible APIs

DOWNLOAD

Open it on the devices you use

iOS

Build from source

Build from source

Android

arm64

Download APK

macOS

Apple silicon

Download .dmg

Windows

x64

Download .exe

Web

Runs in your browser

Open the web app

On first launch, choose a provider in AI service settings, enter its key and test the connection, then start generating. The macOS, Windows and Android apps offer each new version as it comes out.

OPEN SOURCE

Help make Mirra better

Issues and pull requests are welcome: new translations, support for more AI services, better screens and docs, or just telling us what feels off.

$git clone https://github.com/yldm-ai/mirra.git

Report an issue

Found a bug or have an idea? Open it on GitHub Issues.

Issues →

Join the discussion

Share your looks, ask questions, or talk about features in Discussions.

Discussions →

Contribute code

Code, translations and docs are all welcome. Start with the development notes and the architecture.

Development notes →
Thanks to every contributor
houko

FAQ

FAQ

Didn’t find your answer? Ask us on GitHub Discussions.

Is Mirra really free?

Yes. Mirra is open source under MIT, with no fees and no account. If you use a cloud API, your provider bills you; a service you run yourself has no per-call cost.

Is my data uploaded?

Mirra has no servers of its own. API keys stay in your device’s keychain, and photos of your clothes and your looks stay on your device; when you generate, the images you chose and the prompt go straight to the AI service you set up.

Can I use it without a GPU?

Yes. Choose a cloud provider and enter its key. Only running a model yourself needs a GPU that can run it.

Where do the built-in faces, outfits, poses and backgrounds come from?

Model faces come from a CC0-licensed AI portrait library, and the backgrounds are public domain (CC0) photographs that ship with Mirra, with no one in them. Outfits and poses come from third-party reference galleries, which Mirra fetches only when you use them and never bundles or redistributes; see Asset sources & licenses.

Can I use my own or someone else’s photo as the face?

No. Mirra takes no one’s photo: the pose and outfit galleries’ terms forbid using their images to depict real people. Faces come only from the CC0 model library, and the photos you add are of clothes.

Which languages does it speak?

Simplified and Traditional Chinese, English, Japanese, Korean, Spanish, French, German, Brazilian Portuguese, Russian and Italian. It follows your system, or choose one under General in the settings.

How can I contribute?

Read the development notes in the README and ARCHITECTURE.md, then open an issue or a pull request on GitHub. Code, translations and docs are all welcome.