Pretty Alice

• 

  Logo

• 

  Diagram

• 

  Deployment proof

Inspiration for Pretty Alice

I always struggled with doing my own makeup, and my elder sister used to help me with it. After she got married, I suddenly found myself without that guidance. That moment made me realize how helpful it would be to have a companion that could guide me through the process step by step. This personal experience inspired the idea behind Pretty Alice — an AI beauty companion that helps users confidently create their desired looks.

What it does

Pretty Alice is an intelligent beauty assistant designed to guide users through their makeup routine. It provides multiple interactive features including:

• Text Guidance: Step-by-step instructions tailored to the user’s needs.
• Realistic Virtual Try-On: Users can preview makeup looks on their own face, either with or without a reference image.
• Descriptive Face Maps: Clear visual guides showing exactly where each product should be applied.

Pretty Alice also features a Go-Live mode, where users can talk directly with the assistant using voice while the AI sees them in real time. With the Gemini Live API, Pretty Alice becomes a real-time voice coach — guiding users through their makeup routine while their hands are busy with brushes and palettes.

How we built it

After exploring the fundamentals of the ADK and the Gemini Live API, we finalized the concept and began prototyping using Google AI Studio. Our tech stack includes:

Frontend:
Built with React and Tailwind CSS, featuring a "Dark Luxury" aesthetic designed to create a focused and elegant user experience.

Backend:
Powered by Firebase (Authentication and Firestore) to manage user profiles, chat history, and a personalized Beauty Essentials collection.

AI Intelligence:
Utilizes the Gemini 3.1 model suite for advanced reasoning, multimodal understanding, and high-fidelity image editing.

Real-Time Coaching:
Integrated Gemini Live API to enable low-latency multimodal interactions combining voice, text, and visual input.

Image Processing:
Implemented custom image validation and compression pipelines to ensure the AI receives clear facial input and produces accurate visual results.

Challenges we ran into

1. Multimodal Synchronization
   Managing smooth interaction between text, image, and live audio streams was challenging. The Gemini Live API streams audio and processes images simultaneously, but coordinating it with text interactions required additional system logic.

2. Visual Validation
   Preventing the AI from placing makeup on incorrect regions (such as hair or hands) required building a strict visual validator. This system checks facial feature visibility (eyes, lips, skin) before allowing a try-on.

3. Live Tool Coordination
   Managing real-time tool calls without repetitive AI prompts required building a custom processed-call tracking system to keep the conversation natural and efficient.

Accomplishments that we're proud of

Virtual Makeup Kit
Users can add their own makeup products so the AI generates looks using items they already own.

Seamless Virtual Try-On
We achieved high-fidelity makeup edits that preserve the user’s identity while accurately reflecting requested styles.

The "Alice" Personality
We designed Alice to feel like a supportive beauty coach — sophisticated, encouraging, and genuinely helpful.

What we learned

1. Multimodal AI has reached an impressive level where models can simultaneously see, hear, and understand context from different inputs.
2. Building a real-time application taught us a lot about integrating Google Cloud tools into a live product environment.
3. Carefully supervising and testing each AI module is critical to maintaining reliable behavior as features expand.
4. Designing a cohesive system that balances audio, text, and image outputs is essential to creating an assistant that feels unified rather than fragmented.

What's next for Pretty Alice

Community Tutorials
Users will be able to save and share their Alice-assisted makeup looks with a global community.

Smart Mood Boards
Pretty Alice will generate personalized mood boards based on a user’s outfit, jewelry, and event context — helping them choose makeup styles and color palettes that match perfectly.

Built With

• cloudrun
• gemini-adk
• google-ai-studio
• kiro
• vite