​
Overview

FalImageGenService provides high-speed image generation capabilities using fal’s optimized Stable Diffusion XL models. It supports various image sizes, formats, and generation parameters with a focus on fast inference.

​
Installation

To use FalImageGenService, install the required dependencies:

pip install pipecat-ai[fal]

You’ll also need to set up your Fal API key as an environment variable: FAL_KEY

You can obtain a fal API key by signing up at fal.

​
Configuration

​
Constructor Parameters

params
InputParams
required

Generation parameters configuration

aiohttp_session
aiohttp.ClientSession
required

HTTP session for image downloading

model
str
default: "fal-ai/fast-sdxl"

Model identifier

key
str

Fal API key (alternative to environment variable)

​
Input Parameters

class InputParams(BaseModel):
    seed: Optional[int] = None              # Random seed for reproducibility
    num_inference_steps: int = 8            # Number of denoising steps
    num_images: int = 1                     # Number of images to generate
    image_size: Union[str, Dict[str, int]] = "square_hd"  # Image dimensions
    expand_prompt: bool = False             # Enhance prompt automatically
    enable_safety_checker: bool = True      # Filter unsafe content
    format: str = "png"                     # Output image format

​
Supported Image Sizes

Possible enum values: square_hd, square, portrait_4_3, portrait_16_9, landscape_4_3, landscape_16_9

Note: For custom image sizes, you can pass the width and height as an object:

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

See the fal docs for more information.

​
Output Frames

​
URLImageRawFrame

url
string

Generated image URL

image
bytes

Raw image data

size
tuple

Image dimensions (width, height)

format
string

Image format (e.g., ‘PNG’)

​
ErrorFrame

error
string

Error information if generation fails

​
Methods

See the Image Generation base class methods for additional functionality.

​
Usage Example

import aiohttp
from pipecat.services.fal import FalImageGenService

# Configure service
async with aiohttp.ClientSession() as session:
    service = FalImageGenService(
        model="fal-ai/fast-sdxl",
        aiohttp_session=session,
        params=FalImageGenService.InputParams(
            num_inference_steps=8,
            image_size="portrait_hd",
            expand_prompt=True
        )
    )

    # Use in pipeline
    pipeline = Pipeline([
        prompt_input,     # Produces text prompts
        service,          # Generates images
        image_handler     # Handles generated images
    ])

​
Frame Flow

​
Metrics Support

The service collects processing metrics:

  • Generation time
  • Download time
  • API response time
  • Total processing duration

​
Notes

  • Fast inference times with optimized models
  • Supports various image sizes and formats
  • Automatic prompt enhancement option
  • Built-in safety filtering
  • Asynchronous operation
  • Efficient HTTP session management
  • Comprehensive error handling
OpenAI Realtime BetaTavus