Add OpenAI Videos API endpoint support with Sora integration

[Sameerlite]

Title

Add OpenAI Videos API endpoint support with Sora integration

Relevant issues

Fixes LIT-1258

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🆕 New Feature

Changes

Overview

This PR adds comprehensive support for OpenAI's Videos API endpoints, enabling video generation using Sora and Sora-2 models through LiteLLM's unified interface.

Key Features Added

1. Video Endpoints Implementation

• New module: litellm/proxy/openai_videos_endpoints/
  • video_endpoints.py: FastAPI endpoints for all video operations
  • __init__.py: Module initialization
• Endpoints implemented:
  • POST /v1/videos - Create video
  • GET /v1/videos/{video_id} - Retrieve video details
  • GET /v1/videos/{video_id}/content - Download video content
  • DELETE /v1/videos/{video_id} - Delete video
  • GET /v1/videos - List videos

2. Core Video Logic

• New module: litellm/videos/
  • main.py: Core video generation and management functions
  • __init__.py: Module exports
• Functions implemented:
  • create_video() / acreate_video() - Generate videos
  • video_retrieve() / avideo_retrieve() - Get video details
  • video_content() / avideo_content() - Download video content
  • video_delete() / avideo_delete() - Delete videos
  • video_list() / avideo_list() - List videos

3. OpenAI Integration

• Enhanced: litellm/llms/openai/openai.py
  • Added OpenAIVideosAPI class for direct HTTP requests
  • Support for multipart/form-data requests with input_reference parameter
  • Proper error handling and response parsing

4. Cost Calculation Integration

• Enhanced: litellm/cost_calculator.py
  • Added video generation cost calculation logic
  • Support for input_cost_per_video_per_second pricing model
• Enhanced: litellm/litellm_core_utils/llm_cost_calc/utils.py
  • Added cost_per_video_per_second metric support
• Updated: Model cost maps with Sora pricing:
  • sora-2: $0.10 per second
  • sora-2-pro: $0.30 per second

Technical Details

API Compatibility

• Full OpenAI Videos API compatibility
• Support for all Sora model parameters (prompt, model, seconds, size)
• Optional input_reference parameter for image-guided video generation
• Proper multipart/form-data handling for file uploads

Cost Management

• Automatic cost calculation based on video duration and model
• Integration with LiteLLM's existing cost tracking system
• Support for budget limiting and usage monitoring

Usage Example

# Create a video
response = litellm.acreate_video(
    prompt="A cat playing with yarn",
    model="sora-2",
    seconds="4",
    size="720x1280"
)

# With input reference image
response = litellm.acreate_video(
    prompt="Continue this scene",
    model="sora-2-pro", 
    input_reference=("start_frame.jpg", image_bytes, "image/jpeg")
)

This implementation provides a complete, production-ready solution for OpenAI video generation through LiteLLM's unified interface.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
litellm	Error			Oct 13, 2025 7:52am

[ishaan-jaff]

• PR 1 add on SDK level
• Please make sure to follow the same structure as openai/image_generations