🎨 Banana Pro AI (Web & Desktop)

English | 简体中文 | 日本語 | 한국어

Banana Pro AI is a high-performance image generation platform designed for creative professionals. It integrates Gemini and OpenAI standard API capabilities, supporting high-resolution (up to 4K) text-to-image and image-to-image generation, available in both Desktop and Web formats.

[!NOTE] The application supports English internally and allows language switching in the settings.

🆕 v2.8.0 Updates: - 🤖 Dedicated OpenAI Image Generation: New openai-image provider type supporting /v1/images/generations standard API (gpt-image-2 model). - 🎨 Image Card Refactor: Smart thumbnail/full-size switching, improved drag-and-drop, better loading experience.

💡 Recommended: For the best generation experience and cost-effectiveness, we recommend using Yunwu API.

Resolution	Yunwu API Price	Google Official Price (Ref)
1K (1024x1024)	¥0.08 / Image	≈ ¥0.94 / Image
2K (2048x2048)	¥0.08 / Image	≈ ¥0.94 / Image
4K (4096x4096)	¥0.14 / Image	≈ ¥1.68 / Image

---

🌟 Key Features

• 🚀 Extreme Performance: Built with Tauri 2.0 architecture and a high-concurrency Sidecar backend written in Go, ensuring extremely low resource usage.
• 🖼️ 4K Ultra-HD Creation: Deeply optimized Gemini 3.0 model, supporting 4K UHD generation across multiple aspect ratios.
• 🔌 Standard API Compatibility: Supports three provider types: gemini (/v1beta), openai (/v1/chat/completions multimodal), and openai-image (/v1/images/generations) with configurable Base URL and Model ID.
• ⚡ Custom Protocol (asset://): Registered native resource protocol for desktop, bypassing the HTTP stack to increase local image loading speed by 300%.
• 💾 Smart History Management: Built-in local database and persistent caching, supporting task recovery and instant opening of large history records.
• 📸 Precise Image-to-Image: Supports multiple reference images with fine-grained style and composition control.
• 📦 Automated Delivery: Integrated GitHub Actions for automated packaging and releasing on macOS (Intel/M1) and Windows.
• 🧩 Template Market: Prioritizes pulling remote template JSON on startup with automatic fallback to built-in templates.

---

🚀 Functional Details

1. Smart Text-to-Image

• Precise Semantic Understanding: Deep integration with Google Gemini 3.0, capturing fine details, styles, and moods from prompts.
• AI Prompt Optimization: Built-in optimization engine via Gemini / OpenAI standard interfaces.
• Edit History: Supports infinite undo/redo for quick switching between creative ideas.
• Batch Processing: Set up to 100 images for batch generation with background queue processing.
• Real-time Tracking: Clear progress bars and status displays with placeholder cards for each image.

2. Powerful Image-to-Image

• Multi-Ref Support: Add up to 10 reference images to help the AI understand desired composition or style.
• Reverse Prompt Extraction: Click "Extract Prompt" button on reference images to let AI analyze the image and generate detailed prompts. Supports 20+ languages output.
• Flexible Uploads:
  • Click/Drag: Select from local folders or drag-and-drop.
  • Clipboard Support: Paste images directly from the web or chat tools.
• Smart Preprocessing: Automatic compression for oversized images and MD5-based duplicate filtering.

3. Professional Parameter Control

• Aspect Ratios: Preset ratios including 1:1, 16:9, 9:16, 4:3, 2:3.
• Quality Settings: Customizable resolution from 1K to 4K.
• Smart Sizing: Automatically aligns image dimensions to 8-pixel boundaries for optimal model performance.
• Interface Switching: Toggle between Gemini(/v1beta), OpenAI(/v1) multimodal, and OpenAI Image(/v1/images/generations) modes in settings.

4. Advanced UX & Management

• Immersive Preview: Full-screen view with free zooming and dragging.
• High-Density UI: Optimized for productivity with adaptive sidebars and compact info displays.
• Quick Actions:
  • Fast Copy: One-click "Copy Image" button in preview for easy pasting into other apps.
  • Batch Management: Multi-select images for batch saving or deletion.
• Smart Persistence: Remembers sidebar state, window position, and last-used model configurations.

5. Task & History

• Auto Persistence: Real-time saving to local database to prevent data loss.
• Smart Search: Quickly find historical tasks via keywords.
• Stable Connection: Automatically switches between WebSocket and HTTP polling for uninterrupted generation in complex networks.

6. Template Market

• Huge Resource: 900+ high-quality templates across various styles and industries.
• Pull-down Access: Interactive "rope" pull-down to open the market.
• Multi-dim Filtering: Filter by Search, Channel, Material, Industry, or Aspect Ratio.
• PPT Category: Dedicated section for 16:9 templates suitable for presentation materials.
• One-click Reuse: Directly apply templates (replaces current Prompt and reference images).
• Manual Sync: Refresh button to pull latest templates manually.
• Source & Tips: Includes usage tips and clickable source links.
• Ref Requirements: Displays minRefs and note for required reference images.
• Remote Sync: Prioritizes GitHub Raw JSON with local caching.

---

🧩 Template Contribution Guide

Template data is maintained in: - backend/internal/templates/assets/templates.json

Top-level Structure

{
  "meta": {
    "version": "2024.12.01",
    "updated_at": "2024-12-01T12:00:00Z",
    "channels": ["Community", "Social", "Xiaohongshu"],
    "materials": ["Poster", "PPT", "Cover"],
    "industries": ["Education", "Life Services"],
    "ratios": ["1:1", "3:4", "16:9"]
  },
  "items": []
}

Basic Fields (Single Template)

{
  "id": "tpl-001",
  "title": "Cat Meme Template",
  "channels": ["Community", "Entertainment"],
  "materials": ["Poster"],
  "industries": ["Life Services"],
  "ratio": "1:1",
  "preview": "https://.../thumb.jpg",
  "image": "https://.../full.jpg",
  "prompt": "Optional: Template prompt...",
  "prompt_params": "Optional: Prompt usage instructions (reserved)",
  "tips": "Optional: Usage tips/tricks",
  "source": {
    "name": "@Contributor",
    "label": "GitHub",
    "icon": "github",
    "url": "https://example.com/templates/tpl-001"
  },
  "requirements": { "minRefs": 2, "note": "Requires one cat photo as reference" },
  "tags": ["cat", "meme", "funny"]
}

Field Explanation

• requirements.note: Prompt text when reference images are needed.
• requirements.minRefs: Minimum number of reference images required.
• tips: Usage tips/notes (displayed in preview).
• prompt_params: Prompt usage instructions (reserved field, not rendered).
• tags: For searching and aggregation.
• materials: Can include PPT tag (suggested for 16:9) for presentation filtering.
• meta.version / meta.updated_at: For versioning and cache comparison.

source.icon Presets

• github, xhs, wechat, shop, video, print, gov, meme, finance, food, local.

---

🏗️ Technical Architecture

Core Flow

graph TD
    subgraph "Frontend Layer (React + Zustand)"
        UI[User Interface]
        State[Zustand State Management]
        AssetProtocol[asset:// Protocol]
    end

    subgraph "Desktop Container (Tauri 2.0 / Rust)"
        TauriBridge[Rust Bridge]
        IPC[IPC Optimization]
        FS[Local File Access]
    end

    subgraph "Backend Layer (Go Sidecar)"
        GoServer[Gin API Server]
        WorkerPool[Worker Pool]
        GeminiSDK[Google GenAI SDK]
        OpenAIProvider[OpenAI Provider]
        OpenAIImageProvider[OpenAI Image Provider]
        SQLite[(SQLite Storage)]
    end

    UI <--> State
    State <--> IPC
    IPC <--> TauriBridge
    TauriBridge <--> GoServer
    GoServer <--> WorkerPool
    WorkerPool <--> GeminiSDK
    WorkerPool <--> OpenAIProvider
    WorkerPool <--> OpenAIImageProvider
    WorkerPool <--> SQLite
    GeminiSDK <--> |Imagen 3.0| Cloud[Google AI Cloud]
    OpenAIProvider <--> |/v1/chat/completions| OpenAI[OpenAI Compatible API]
    OpenAIImageProvider <--> |/v1/images/generations| OpenAIImg[OpenAI Image API]
    GoServer -.-> |Save Images| FS
    FS -.-> |Map Resource| AssetProtocol
    AssetProtocol -.-> |Fast Display| UI

The project uses a "three-layer architecture" to balance performance and scalability: 1. Frontend (React + Zustand): Handles responsive UI and state management. 2. Desktop Container (Tauri): Acts as a Rust bridge for window control and local resource access. 3. Inference Engine (Go Sidecar): Communicates with AI providers (Gemini, OpenAI, OpenAI-Image) and manages task pools.

Core Optimizations

• IPC Load Optimization: Only file paths are passed between frontend and backend; large binary data is read directly via the asset:// protocol.
• Lifecycle Management: Automatically cleans up Go sidecar processes when Tauri exits.

---

📂 Project Structure

├── backend/            # Go Backend (Sidecar)
│   ├── cmd/server/     # Entry point
│   └── internal/       # Core logic (Gemini, Worker, DB)
├── desktop/            # Tauri Desktop Project (React + Rust)
│   ├── src/            # Frontend logic
│   └── src-tauri/      # Rust & System permissions
├── frontend/           # Independent Web Frontend (Reference)
└── assets/             # Presentation resources

---

💻 Developer Guide

1. Prerequisites

• Go: 1.21+
• Node.js: 18+
• Rust: 1.75+ (Required for Tauri)

🍎 macOS Permission Fix

If you encounter a "Damaged" error on macOS due to Gatekeeper, run:

sudo xattr -r -d com.apple.quarantine "/Applications/Banana Pro AI.app"

2. Backend Setup

cd backend
# Configure config.yaml with your API Key
go run cmd/server/main.go

Or use Makefile:

make build    # Compile backend
make run      # Run backend

3. Desktop Setup

cd desktop
npm install
npm run tauri dev

4. Web Frontend Setup

cd frontend
npm install
npm run dev

5. Automated Build (GitHub Actions)

Push a version tag to trigger CI:

git tag v2.8.0
git push origin v2.8.0

6. Auto Updater

Integrated Tauri Updater for one-click updates. 1. Generate keys: npm run tauri signer generate -- -w ~/.tauri/banana-updater.key 2. Add public key to tauri.conf.json. 3. Configure GitHub Secrets for CI.

---

⚙️ Core Configuration

Item	Description
AI Provider	gemini (/v1beta), openai (/v1/chat/completions), or openai-image (/v1/images/generations). Each uses its own Base URL and model.
API Base / Key	Standard OpenAI format compatibility.
Image Model	Primary model for image generation (e.g., gemini-2.0-flash-exp, gpt-4o, gpt-image-2).
Vision Model	Model for reverse prompt extraction. Inherits Image Model's Base URL and API Key by default.
Chat Model	Model for prompt optimization.
Storage Dir	Default to system AppData (Win) or Application Support (Mac).
Templates Remote URL	Remote template JSON URL (defaults to GitHub Raw).
asset://	Custom protocol for fast local image access.

---

🐳 Docker Deployment (Web)

Only for Backend + Web Frontend deployment.

Quick Start

# 1. Copy environment template and configure API Key
cp .env.example .env
nano .env  # Add your GEMINI_API_KEY or OPENAI_API_KEY

# 2. Start services (must use docker compose)
docker compose -p banana-pro up -d

# 3. Access the application
# Browser: http://localhost:8090

Detailed Documentation

For complete deployment guide, configuration, and troubleshooting, see: DOCKER_DEPLOY.md

Key Features

• 🐳 Multi-stage Build: Frontend (Node.js) + Backend (Go) + Runtime (Alpine + Nginx)
• 🚀 Environment Auto-Detection: Backend automatically detects Docker and listens on 0.0.0.0 (Tauri uses 127.0.0.1)
• 💾 Data Persistence: Images and database automatically mounted to ./data/storage
• 🔄 Health Check: Built-in health endpoint with automatic restart
• 🇨🇳 Mirror Support: Configurable China mirror sources via Build Args

---

🤝 Contribution & Feedback

We welcome all forms of contribution!

• Bug Reports: Use GitHub Issues with detailed reproduction steps.
• PRs: Follow existing style and test thoroughly before submitting.

---

📄 License

This project is licensed under the MIT License.

---

📈 Star His