MCP 服务器

fal-hidream-i1-full MCP Server

Enables high-quality AI image generation using the fal-ai/hidream-i1-full model with support for synchronous, streaming, and queue-based generation, custom image sizing, LoRA weights, and automatic local image downloads.

README

fal-ai/hidream-i1-full MCP Server

A Model Context Protocol (MCP) server that provides access to the fal-ai/hidream-i1-full image generation model. This server allows you to generate high-quality images using advanced AI technology through the fal.ai platform.

Features

High-Quality Image Generation: Generate stunning images using the fal-ai/hidream-i1-full model
Multiple Generation Methods: Support for synchronous, streaming, and queue-based generation
Flexible Image Sizing: Support for predefined sizes and custom dimensions
Advanced Parameters: Control over inference steps, guidance scale, safety checker, and more
LoRA Support: Apply custom LoRA weights for specialized image styles
Local Image Download: Automatically downloads generated images to local storage
Queue Management: Submit long-running requests and check their status
Webhook Support: Optional webhook notifications for completed requests

Installation

Clone this repository:

git clone https://github.com/PierrunoYT/fal-hidream-i1-full-mcp-server.git
cd fal-hidream-i1-full-mcp-server

Install dependencies:

npm install

Build the project:

npm run build

Configuration

Environment Variables

Set your fal.ai API key as an environment variable:

export FAL_KEY="your_fal_api_key_here"

You can get your API key from fal.ai.

MCP Client Configuration

Add this server to your MCP client configuration. For example, in Claude Desktop's config file:

{
  "mcpServers": {
    "fal-hidream-i1-full": {
      "command": "node",
      "args": ["/path/to/fal-hidream-i1-full-mcp-server/build/index.js"],
      "env": {
        "FAL_KEY": "your_fal_api_key_here"
      }
    }
  }
}

Available Tools

1. `hidream_i1_full_generate`

Generate images using the standard synchronous method.

Parameters:

prompt (required): Text description of the image to generate
negative_prompt (optional): What you don't want in the image
image_size (optional): Predefined size or custom {width, height} object
num_inference_steps (optional): Number of inference steps (1-100, default: 50)
seed (optional): Random seed for reproducible results
guidance_scale (optional): CFG scale (1-20, default: 5)
sync_mode (optional): Wait for completion (default: true)
num_images (optional): Number of images to generate (1-4, default: 1)
enable_safety_checker (optional): Enable safety filtering (default: true)
output_format (optional): "jpeg" or "png" (default: "jpeg")
loras (optional): Array of LoRA weights to apply

Example:

{
  "prompt": "a cat holding a skateboard which has 'fal' written on it in red spray paint",
  "image_size": {"width": 1024, "height": 1024},
  "num_inference_steps": 50,
  "guidance_scale": 7.5
}

2. `hidream_i1_full_generate_stream`

Generate images using streaming for real-time progress updates.

Parameters: Same as hidream_i1_full_generate

3. `hidream_i1_full_generate_queue`

Submit a long-running image generation request to the queue.

Parameters: Same as hidream_i1_full_generate plus:

webhook_url (optional): URL for webhook notifications

Returns: A request ID for tracking the job

4. `hidream_i1_full_queue_status`

Check the status of a queued request.

Parameters:

request_id (required): The request ID from queue submission
logs (optional): Include logs in response (default: true)

5. `hidream_i1_full_queue_result`

Get the result of a completed queued request.

Parameters:

request_id (required): The request ID from queue submission

Image Sizes

Predefined Sizes

square_hd: High-definition square
square: Standard square
portrait_4_3: Portrait 4:3 aspect ratio
portrait_16_9: Portrait 16:9 aspect ratio
landscape_4_3: Landscape 4:3 aspect ratio
landscape_16_9: Landscape 16:9 aspect ratio

Custom Sizes

You can also specify custom dimensions:

{
  "image_size": {
    "width": 1280,
    "height": 720
  }
}

LoRA Support

Apply custom LoRA weights for specialized styles:

{
  "loras": [
    {
      "path": "https://example.com/lora-weights.safetensors",
      "scale": 1.0,
      "weight_name": "optional_weight_name"
    }
  ]
}