Replicate AI Image Generator

Web application for generating images using AI models with automatic translation of prompts into English.

Automatic installation

1. Clone the repository:

git clone https://github.com/zdenek-stursa/replicator.git
cd replicator

2. Create and activate a virtual environment:

sudo ./app.sh --install

3. Create a .env file based on the .env.example template and fill in the necessary API keys:

cp .env.example .env

• Important: In the .env file, set the REPLICATE_MODELS variable with a list of models from Replicate that you want to use. Separate models with a comma (e.g., owner/model-name:version,another/model:version). See .env.example for the format.
• Note: The application uses liteLLM for flexible LLM provider support. Configure your API key for your preferred provider (OpenAI, Anthropic, xAI, etc.) and set the LLM_MODEL variable accordingly.

Manual installation

1. Clone the repository:

git clone https://github.com/zdenek-stursa/replicator.git
cd replicator

2. Create and activate a virtual environment:

python3 -m venv venv
source venv/bin/activate  # On Windows: .\venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Create a .env file based on the .env.example template and fill in the necessary API keys:

cp .env.example .env

• Important: In the .env file, set the REPLICATE_MODELS variable with a list of models from Replicate that you want to use. Separate models with a comma (e.g., owner/model-name:version,another/model:version). See .env.example for the format.
• Note: The application uses liteLLM for flexible LLM provider support. Configure your API key for your preferred provider (OpenAI, Anthropic, xAI, etc.) and set the LLM_MODEL variable accordingly.

Model Configuration

The application loads the list of available models for generation from the REPLICATE_MODELS environment variable defined in the .env file.

The format is a comma-separated list of model identifiers:

REPLICATE_MODELS=model_identifier_1,model_identifier_2,...

The identifier should be in the format owner/model-name:version (the version is optional; if not provided, the latest version will be used).

The user interface then dynamically loads the parameters for each selected model directly from the Replicate API and displays the corresponding form.

Running for Development

./app.sh --debug # Or: FLASK_ENV=development python app.py

Automatic Production Deployment

1. Run the application using Gunicorn:

./app.sh --production

Manual Production Deployment

1. Set production variables in .env:

FLASK_ENV=production
SECRET_KEY=your-secure-secret-key
RATELIMIT_STORAGE_URL=redis://localhost:6379/0

2. Run the application using Gunicorn:

gunicorn --bind 0.0.0.0:5000 --workers 4 app:app

Features

• Image Generation: Generate images using various AI models from Replicate (configurable via .env)
• Model Selection: Choose from multiple pre-configured models with dynamic parameter loading
• Multi-LLM Support: Flexible LLM integration via liteLLM library
  • Automatic translation of prompts into English
  • Prompt enhancement for better image generation
  • Support for multiple providers: OpenAI, Anthropic, xAI, Groq, Mistral, and more
  • Easy model switching via environment configuration
• Image Gallery: View previously generated images with metadata and PhotoSwipe lightbox
• Image Format Conversion: Download images in multiple formats (WebP original, JPG 90% quality, PNG without transparency) with mobile-optimized UI
• Aspect Ratio Support: Select from predefined aspect ratios or set custom dimensions
• Rate Limiting: Built-in API protection with configurable limits
• Security: Security headers, CORS protection, and secure cookie settings
• Responsive Design: Works on desktop and mobile devices
• Comprehensive Logging: File logging with rotation for debugging and monitoring
• Modular Frontend Architecture: ES6 modules for better code organization and maintainability

LLM Configuration

The application uses liteLLM for flexible LLM provider support. Configure your preferred model via the LLM_MODEL environment variable:

# OpenAI models (default)
LLM_MODEL=gpt-4
LLM_MODEL=gpt-4.1
LLM_MODEL=gpt-4.1-mini
LLM_MODEL=gpt-4.1-nano
LLM_MODEL=gpt-4-turbo
LLM_MODEL=gpt-4o
LLM_MODEL=gpt-4o-mini
LLM_MODEL=gpt-3.5-turbo
LLM_MODEL=o3
LLM_MODEL=o3-mini
LLM_MODEL=o4-mini
LLM_MODEL=o1-mini
LLM_MODEL=o1-preview

# Anthropic Claude
LLM_MODEL=claude-3-opus-20240229
LLM_MODEL=claude-3-sonnet-20240229

# xAI Grok
LLM_MODEL=xai/grok-beta

# Groq
LLM_MODEL=groq/llama3-8b-8192

# Mistral
LLM_MODEL=mistral/mistral-7b-instruct

Note: Special model versions (e.g., gpt-4-1-2025-04-14) are automatically mapped to standard names for compatibility.

Image Format Conversion

The application supports on-demand image format conversion with automatic cleanup:

Supported Formats

• WebP (Original): Direct download of the original image from Replicate
• JPEG: Converted with 90% quality compression, transparency removed
• PNG: Converted without transparency support

Features

• On-demand conversion: Images are converted only when requested
• Temporary file management: Converted files are automatically cleaned up after 2 hours
• Rate limiting: 30 conversion requests per minute
• Error handling: Comprehensive error handling with user feedback
• Mobile-optimized UI: Touch-friendly dropdown menu with backdrop and improved positioning

API Endpoints

• GET /api/convert/<image_id>/jpg - Convert and download as JPEG
• GET /api/convert/<image_id>/png - Convert and download as PNG
• GET /api/temp-files-info - Get temporary files statistics (debugging)

Image Gallery with PhotoSwipe

The application uses PhotoSwipe v5.4.4 for professional image viewing experience:

Features

• Keyboard Navigation: Arrow keys for navigation, Escape to close
• Touch/Swipe Support: Full touch gesture support on mobile devices
• Zoom Functionality: Zoom up to 3x including 1:1 pixel viewing
• Proper Aspect Ratios: Automatic detection of image dimensions to prevent distortion
• Responsive Design: Optimized for both desktop and mobile viewing
• Smooth Animations: Professional fade transitions and zoom effects

Technical Implementation

• Uses PhotoSwipe UMD version loaded from CDN
• Automatic image dimension detection using Image objects
• Asynchronous loading for optimal performance
• Fallback to new tab if PhotoSwipe fails to load

Rate Limits

• Image generation: 5 requests/minute
• Prompt enhancement: 10 requests/minute
• Gallery listing: 30 requests/minute
• Image download: 60 requests/minute
• Image conversion: 30 requests/minute

Frontend Architecture

The frontend uses a modular ES6 architecture for better maintainability and code organization:

JavaScript Modules

• constants.js - Application constants and configuration
• storage.js - localStorage management and form state persistence
• ui.js - UI utilities (error handling, loading states, notifications)
• gallery.js - Image gallery management and pagination
• form-generator.js - Dynamic form generation and aspect ratio handling
• api-client.js - API communication and data fetching
• photoswipe-gallery.js - PhotoSwipe lightbox integration with automatic image dimension detection
• mobile-dropdown.js - Mobile-optimized dropdown menu enhancements for touch devices
• main.js - Application initialization and event handling

Benefits

• Separation of Concerns: Each module has a specific responsibility
• Maintainability: Smaller, focused files are easier to understand and modify
• Reusability: Modules can be imported and used independently
• Testing: Individual modules can be tested in isolation
• Modern Standards: Uses ES6 import/export syntax

Security

• Rate limiting
• Security headers
• CORS protection
• Secure cookie settings