AI Image Generation CLI Tools

A unified command-line interface for AI image generation services: Venice.ai, Fal.ai, Wavespeed.ai, and Replicate.ai. Generate images, videos, and 3D models using state-of-the-art AI models.

Installation

npm install -g flux-client

Available Commands

Venice AI (venice)

Generate images using Venice AI's models with extensive customization options.

# Basic usage
venice --prompt "A futuristic cityscape at dusk"

# Advanced options
venice --prompt "A serene landscape" --format wide --lora photographic
venice --prompt "A cyberpunk scene" --steps 25 --cfg-scale 2 --seed 42
venice --prompt "Portrait" --model qwen-image --variants 4 --output-format png

Options:

• --prompt <text>: Text description of the image to generate
• --negative-prompt <text>: Describe what to avoid in the image
• --model <modelKey>: Choose AI model (run venice --help to see available models)
• --format <formatKey>: Image size preset (square, portrait, landscape, wide, tall)
• --width <number>: Custom image width (default: 1024)
• --height <number>: Custom image height (default: 1024)
• --lora <preset>: Apply a Venice LoRA/style preset (Anime, Photographic, Cinematic, Fantasy Art, etc.)
• --steps <number>: Number of inference steps (default: 20)
• --cfg-scale <number>: Guidance scale (default: 2)
• --seed <number>: Random seed for reproducible results
• --output-format <format>: Output format (jpeg, png, webp - default: webp)
• --variants <number>: Generate multiple variants (1-4, only with --return-binary false)
• --lora-strength <number>: LoRA strength (0-100)
• --embed-exif-metadata: Embed generation parameters in EXIF metadata
• --hide-watermark: Hide watermark in generated image
• --return-binary: Return image as binary data (default: true)
• --debug: Enable debug mode

Note: Images are automatically constrained to dimensions divisible by 16 (model requirement), with a maximum of 1280x1280.

Venice Models (venice-models)

Fetch and update the latest available Venice AI models from the API.

venice-models  # Updates venice/models.json with latest models

---

Fal AI (falflux)

Generate images, videos, and 3D models using Fal AI's extensive model library. Supports LoRAs, model discovery, and advanced features.

Basic Usage

# Basic image generation (default model: krea-lora)
falflux --prompt "A futuristic cityscape at dusk"

# Using specific models
falflux --model pro --prompt "Photorealistic portrait"
falflux --model ultra --prompt "High quality landscape"
falflux --model dev --prompt "Creative artwork"

LoRA Support

Apply artistic styles using Low-Rank Adaptation (LoRA) models:

# Single LoRA
falflux --lora disney --prompt "An enchanted forest"
falflux --lora cinematic --prompt "A dramatic sunset" --format wide

# Multiple LoRAs (comma-separated)
falflux --lora disney,lucid --prompt "A magical landscape"
falflux --lora anime,niji --prompt "Anime character in mystical setting"

Available LoRAs: disney, lucid, retrowave, incase, eldritch, details, mj, fantasy, poly, cinematic, anime, anime-flat, anime-portrait, niji, fantasy-core, goofy, psychedelic, neurocore, anime-realistic

Run falflux --help for the complete list.

Video Generation

Generate videos from images or text:

# Image-to-video (requires input image)
falflux --model image_to_video --image-url "./photo.jpg" --prompt "Camera slowly pans right"

# With duration control
falflux --model image_to_video --image-url "./photo.jpg" --prompt "Zoom in slowly" --duration 10

# Video models: image_to_video, wan-25, hunyuan

Image-to-Image

Transform existing images:

falflux --model krea-i2i --image-url "./input.jpg" --prompt "Transform into watercolor style"
falflux --model krea-i2i --image-url "./photo.jpg" --prompt "Make it look like oil painting" --strength 0.7

Model Discovery

Explore the extensive model library:

# List all available categories
falflux --list-categories

# List all models (grouped by category)
falflux --list-models

# List models in a specific category
falflux --list-models --category text-to-image
falflux --list-models --category image-to-video

# Search for models by keyword
falflux --search "anime"
falflux --search "video"
falflux --search "3d"

# Get detailed information about a model
falflux --model-info pro
falflux --model-info krea-lora

Options:

• --prompt <text>: Text prompt for generation
• --model <modelKey>: Model to use (default: krea-lora)
• --format <formatKey>: Image size (square, square_hd, portrait, landscape, wide, tall)
• --lora <keys>: Apply LoRAs (comma-separated)
• --seed <number>: Random seed for reproducibility
• --scale <number>: Guidance scale (default: 2.7)
• --strength <number>: Strength for image-to-image (0.01-1.0)
• --num-images <number>: Number of images to generate (1-4, default: 1)
• --image-url <path>: Input image (URL or local file path) for image-to-video/image-to-image
• --duration <seconds>: Video duration (5 or 10 seconds, default: 5)
• --out: Save to current directory instead of default
• --debug: Enable debug mode
• --list-models: List all available models
• --list-categories: List all model categories
• --search <term>: Search models by keyword
• --model-info <key>: Show detailed model information

Note: Local image files are automatically uploaded to Fal storage when used with --image-url.

---

Wavespeed AI (wavespeed)

Generate high-quality images using Wavespeed AI's fast and powerful models, including FLUX.2, Seedream, and more.

Basic Usage

# Basic image generation (default model: z-image-turbo)
wavespeed --prompt "A futuristic cityscape at dusk"

# Using specific models
wavespeed --model flux2 --prompt "Photorealistic portrait"
wavespeed --model seedream --prompt "High quality landscape"
wavespeed --model grok --prompt "Product photography shot"

Available Models

• flux-2-flex, flux2, flex - FLUX.2 [flex]: Fast, flexible text-to-image with enhanced realism (1536x1536 max)
• z-image-turbo, z-image, turbo - Z-Image-Turbo: 6B parameter model, photorealistic in sub-second time (1536x1536 max, default)
• seedream-v4.5, seedream, v4.5 - Seedream v4.5: Latest version by ByteDance with improved quality (4096x4096 max)
• seedream-v4, v4 - Seedream v4: High-fidelity image generation (4096x4096 max)
• seedream-v3.1, v3.1 - Seedream v3.1: Strong style fidelity and rich detail (2048x2048 max)
• wan-2.5, wan2.5, wan - WAN 2.5: Alibaba text-to-image model (1440x1440 max)
• grok-2-image, grok2, grok - Grok 2 Image: xAI's photorealistic image generation (1536x1536 max)

Prompt Optimization

Wavespeed includes a built-in prompt optimizer that enhances your prompts before generation:

# Basic optimization
wavespeed --prompt "woman walking" --optimize

# Optimization for video prompts
wavespeed --prompt "city scene" --optimize --optimize-mode video

# Style-specific optimization
wavespeed --prompt "portrait shot" --optimize --optimize-style photographic
wavespeed --prompt "fantasy art" --optimize --optimize-style artistic
wavespeed --prompt "anime character" --optimize --optimize-style anime

# With reference image
wavespeed --prompt "similar style" --optimize --optimize-image https://example.com/reference.jpg

Optimization Modes: image (default), video Optimization Styles: default, artistic, photographic, technical, anime, realistic

Advanced Features

# Custom image sizes
wavespeed --prompt "Mountain landscape" --format 1920*1080
wavespeed --prompt "Square image" --format square_hd

# Multiple generations
wavespeed --prompt "A magical landscape" --count 4

# With seed for reproducibility
wavespeed --prompt "A magical landscape" --seed 12345

# Synchronous mode (wait for result in single response)
wavespeed --prompt "Quick test" --sync

# Save to current directory
wavespeed --prompt "Local save" --out

# Debug mode
wavespeed --prompt "Test" --debug

Options:

• --prompt <text>: Text prompt for generation
• --model <modelKey>: Model to use (default: z-image-turbo)
• --format <formatKey>: Image size preset (square, portrait, landscape, wide, tall) or custom dimensions
• --seed <number>: Random seed for reproducibility
• --count <number>: Number of generations (default: 1)
• --optimize: Enable prompt optimization
• --optimize-mode <mode>: Optimization mode (image, video)
• --optimize-style <style>: Optimization style (default, artistic, photographic, technical, anime, realistic)
• --optimize-image <url>: Reference image URL for optimization
• --sync: Enable synchronous mode
• --out: Save to current directory instead of default
• --debug: Enable debug mode

Note: Images are automatically constrained to model-specific maximum dimensions while preserving aspect ratio.

---

Replicate (repflux)

Download and manage existing predictions from your Replicate account.

# Download all predictions from your account
repflux

# Get a specific prediction by ID
node replicate/get.js <prediction-id>

Note: repflux does not generate new images. It downloads existing predictions from your Replicate account. Use Replicate's web interface or SDK to create new predictions.

Environment Variables

All services require their respective API credentials to be set as environment variables:

export VENICE_API_TOKEN="your-venice-api-token"
export FAL_KEY="your-fal-api-key"
export WAVESPEED_KEY="your-wavespeed-api-key"
export REPLICATE_API_TOKEN="your-replicate-api-token"

Add these to your ~/.bashrc, ~/.zshrc, or equivalent shell configuration file for persistence.

Output Directories

Generated media is automatically saved to the following locations:

• Venice AI: ./images/venice/ (automatically created)
• Fal AI: ./images/ (or $FAL_PATH if set)
• Replicate: ./output/ (or $REPLICATE_OUTPUT_DIR if set)

File naming convention: <source>_<timestamp>.<ext> or extracted from URL for downloads.

Prompt Files

All services support reading prompts from a prompt.txt file in the current directory:

# Create a prompt file
echo "A serene mountain landscape at sunset" > prompt.txt

# Generate without --prompt flag (uses prompt.txt)
venice
falflux --model pro

This approach is useful for:

• Avoiding shell escaping issues with complex prompts
• Batch workflows
• Long prompts that exceed command-line limits

Additional Features

Model Discovery (Fal)

Use the built-in model discovery commands to explore available models:

falflux --list-categories  # See all categories
falflux --search "anime"   # Find anime models
falflux --model-info dev   # Get details about a model

Debug Mode

Enable verbose logging with the --debug flag:

venice --prompt "test" --debug
falflux --prompt "test" --debug

Development

See CLAUDE.md for detailed development documentation, architecture details, and contribution guidelines.