AI-Powered 3D Model Generation — Android App

Generate, refine, and visualize three-dimensional models from text descriptions and photos using AI — directly on your Android device.

[PASTE IMAGE]

---

Table of Contents

• Overview
• Screenshots
• Tech Stack
• Architecture
• AI Description Flow
• Core Features
• Setup & Configuration
• Supported Formats
• Testing

---

Overview

This Android application lets users create accurate 3D models through a guided AI conversation or by uploading a photo. Models can then be explored in an interactive 3D scene or placed into the real world via Augmented Reality.

Key capabilities:

• AI-assisted prompt refinement via GPT before generation
• Text-to-3D and image/photo-to-3D via Meshy AI
• Interactive 3D viewer (Filament) and AR placement (ARCore)
• Model library with category filtering, history, and deletion
• Cloud-synced user data via Firebase + Firestore

---

Screenshots

Sign In

User Cabinet

Text-to-3D Generation via AI Chat

Image-to-3D — Photo Input

Model Library

3D Viewer

AR Mode

---

Tech Stack

Tool / Service	Purpose
Meshy AI	3D model generation API — text-to-3D and image-to-3D
OpenAI GPT API	Conversational assistant for refining model prompts
Firebase Authentication	User sign-in and session management
Firestore	NoSQL cloud database for user data and model metadata
Amazon S3	Object storage for generated 3D model assets
Retrofit	HTTP client — maps REST endpoints to Kotlin interfaces
Hilt	Dependency injection
WorkManager	Background task scheduling for long-running generation jobs
Filament	Real-time 3D rendering engine
ARCore	Augmented Reality placement and tracking

---

Architecture

The app follows Clean Architecture with three layers: UI → Domain → Data.

UI Layer        ViewModel  ←→  UseCase
                                  ↕
Domain Layer             Repository (interface)
                                  ↕
Data Layer       RepositoryImpl  →  Remote API / Firestore / S3

Example flow — text-to-3D:

ChatScreen → ChatViewModel → SendMessageUseCase → GPT RepositoryImpl → OpenAI API
                                    ↓
                          GenerateModelFromTextUseCase
                                    ↓
                     WorkManagerMeshyRepo → GetTextModelIdWorker
                                              → GetTextModelWorker
                                              → SaveModelWorker

Key component roles:

• ViewModel — holds and exposes UI state; survives configuration changes
• UseCase — single-responsibility business logic unit
• Repository — abstracts data sources behind a stable interface
• Workers — run generation polling in the background via WorkManager

---

AI Model Description Flow

Before submitting a prompt to Meshy, the user refines it through a structured GPT conversation.

Conversation Rules

Trigger	Behavior
Regular user message	GPT asks targeted follow-up questions (color, size, shape, material, style, quality)
Message contains NEXT	GPT continues asking refinement questions
Message contains END	GPT immediately outputs the final description

Final response always begins with: FINAL object is …

Style Options

fantasy · cartoon · sci-fi · futurist · realistic · ancient · elegant · ultra realistic · trending on artstation · masterpiece · cinema 4d · unreal engine · octane render

Quality Options

highly detailed · high resolution · highest quality · best quality · 4K · 8K · HDR · studio quality

---

Core Features

3D Model Creation

• Text-to-3D — describe an object through the GPT assistant; the refined prompt is sent to Meshy AI for generation
• Image/Photo-to-3D — upload a photo from the gallery or capture with camera; Meshy AI reconstructs a 3D model

Model Management

• Browse all generated models in a list with category filter chips
• View model metadata and generation history
• Delete unwanted models

3D Viewer

• Real-time interactive rendering powered by Filament
• Rotate, zoom, and pan the model in a full 3D scene

AR Mode

• Place any generated model into the real world via ARCore
• Walk around and inspect the model at true scale

Model Refinement

• Post-generation improvement workflow to re-submit or adjust prompts

User Authentication & Sync

• Sign in with Firebase Authentication
• All model metadata and user preferences synced with Firestore
• Model assets stored in Amazon S3

---

Setup & Configuration

Note: API keys and service credentials are required before building.

1. Meshy AI — add your API key to local.properties:

   MESHY_API_KEY=your_key_here
   

2. OpenAI — add your API key:

   OPENAI_API_KEY=your_key_here
   

3. Firebase — place your google-services.json in app/
4. Amazon S3 — configure bucket name and credentials in local.properties or environment variables
5. Sync Gradle and build the project

---

Supported Formats

Models can be exported and loaded in the following formats:

Format	Use case
GLB	Filament 3D viewer, general-purpose
FBX	DCC tools (Blender, Maya, etc.)
USDZ	iOS AR Quick Look (cross-platform sharing)

---

Testing

Manual testing was conducted using structured use-case scenarios covering all core user flows. See the full test documentation:

TESTING.md — Use Cases 3.1 – 3.8 (авторизація, реєстрація, створення моделі, перегляд, видалення, налаштування, зміна паролю)