Implementation:BerriAI Litellm Veo Video Generation Example

Attribute	Value
Sources	cookbook/veo_video_generation.py
Domains	Video Generation, Google Veo, Pass-Through, Cookbook
Last Updated	2026-02-15 16:00 GMT

Overview

veo_video_generation is a cookbook example demonstrating a complete workflow for generating videos using Google's Veo model through the LiteLLM proxy's Gemini pass-through endpoint, including submission, polling, and download.

Description

This script implements the VeoVideoGenerator class that provides a full video generation pipeline through the LiteLLM proxy:

• generate_video - Initiates video generation by sending a text prompt to the Veo predictLongRunning endpoint via the LiteLLM proxy's Gemini pass-through (/gemini/v1beta). Returns the operation name for status polling.

• wait_for_completion - Polls the operation status endpoint with exponential backoff (starting at 10 seconds, capping at 30 seconds), checking for completion up to a configurable max wait time (default: 600 seconds). Extracts the video URI from the nested response structure on success.

• download_video - Downloads the generated video file via the LiteLLM proxy, converting Google's file URI (e.g., files/abc123) to the proxy's download URL format (files/abc123:download?alt=media). Streams the download with progress reporting.

• generate_and_download - Orchestrates the complete workflow: generate, poll, and download, with auto-generated filenames based on the prompt and timestamp.

The script is configurable via LITELLM_BASE_URL and LITELLM_API_KEY environment variables and includes example prompts for testing.

Usage

Run this script as a standalone example when testing or demonstrating video generation through the LiteLLM proxy with Google's Veo model. Requires a running LiteLLM proxy with Google AI Studio pass-through configured and a Google AI Studio API key with Veo access.

Code Reference

Source Location

cookbook/veo_video_generation.py

Signature

class VeoVideoGenerator:
    def __init__(self, base_url: str = "http://localhost:4000/gemini/v1beta", api_key: str = "sk-1234")
    def generate_video(self, prompt: str) -> Optional[str]
    def wait_for_completion(self, operation_name: str, max_wait_time: int = 600) -> Optional[str]
    def download_video(self, video_uri: str, output_filename: str = "generated_video.mp4") -> bool
    def generate_and_download(self, prompt: str, output_filename: str = None) -> bool

def main()

Import

# This is a standalone cookbook script, run directly:
# python cookbook/veo_video_generation.py

I/O Contract

Inputs

Parameter	Type	Description
base_url	str	Base URL for the LiteLLM proxy with Gemini pass-through. Defaults to "http://localhost:4000/gemini/v1beta".
api_key	str	API key for LiteLLM proxy authentication.
prompt	str	Text description of the video to generate.
operation_name	str	The operation name returned by generate_video for status polling.
max_wait_time	int	Maximum polling time in seconds. Defaults to 600.
video_uri	str	The URI of the generated video for download.
output_filename	Optional[str]	Local filename for the downloaded video.

Outputs

Method	Return Type	Description
generate_video	Optional[str]	The operation name if generation started, None on failure.
wait_for_completion	Optional[str]	The video URI if complete, None on timeout or error.
download_video	bool	True if download succeeded, False otherwise.
generate_and_download	bool	True if the full workflow succeeded, False otherwise.

Usage Examples

import os

# Set environment variables
os.environ["LITELLM_BASE_URL"] = "http://localhost:4000/gemini/v1beta"
os.environ["LITELLM_API_KEY"] = "sk-1234"

# Run from the command line:
# python cookbook/veo_video_generation.py

# Or use programmatically:
from cookbook.veo_video_generation import VeoVideoGenerator

generator = VeoVideoGenerator(
    base_url="http://localhost:4000/gemini/v1beta",
    api_key="sk-1234",
)

# Generate and download a video
success = generator.generate_and_download(
    prompt="A cat playing with a ball of yarn in a sunny garden",
    output_filename="cat_video.mp4",
)

# Or step-by-step:
operation = generator.generate_video("Ocean waves at sunset")
if operation:
    video_uri = generator.wait_for_completion(operation)
    if video_uri:
        generator.download_video(video_uri, "ocean.mp4")

Related Pages

• BerriAI_Litellm_Nova_Sonic_Realtime_Example - Another cookbook example using LiteLLM proxy pass-through
• BerriAI_Litellm_Passthrough_Utils - Utilities for pass-through endpoint handling