SamurAIGPT/Seedance-2.5-API

Python wrapper for ByteDance's Seedance 2.5 API — Text-to-Video, Image-to-Video, realistic human faces, native 4K, consistent character generation.

What it solves

This project provides a comprehensive Python wrapper for the Seedance 2.5 API (developed by ByteDance), allowing developers to programmatically generate high-fidelity AI videos. It simplifies the process of creating cinematic clips from text prompts, static images, or a combination of multimodal references, while specifically addressing the challenge of maintaining consistent human faces and character identity across different scenes.

How it works

The library acts as an SDK for the Seedance 2.5 model delivered via MuAPI. It provides high-level Python methods to interact with various endpoints, including:

  • Text-to-Video (T2V) and Image-to-Video (I2V) generation.
  • Omni-Reference: Conditioning video generation on up to 30 images, 10 videos, and 10 audio clips simultaneously.
  • Character Consistency: A specialized workflow where users can create a 4K character sheet from 1-3 reference photos and then reference that character using a unique ID (@character:<id>) or by using the sheet as an anchor image for higher fidelity.
  • Video Editing: Modifying existing videos using text and image prompts.
  • Synced Audio: Generating matched voice, sound effects, and background music.

Who it’s for

  • AI Video Creators: Those looking to produce high-quality cinematic clips with realistic human faces.
  • Developers: Those building AI video applications or SaaS platforms that require a robust Python interface for ByteDance's video generation models.
  • AI Agents: The project includes an MCP (Model Context Protocol) server, enabling AI models like Claude or Cursor to trigger video generation tools directly.

Highlights

  • Extended Clip Length: Supports clips up to 30 seconds long.
  • Realistic Human Faces: Specialized focus on facial fidelity and identity consistency to avoid the "uncanny valley."
  • Multimodal Conditioning: Ability to use images, videos, and audio as references in a single request.
  • ** idée Fixed-Camera Control**: A camera_fixed option to bias the model toward static shots.
  • High-Fidelity Formats: Support for .mov (yuv444p + PCM audio) to prevent color drift and audio desync during repeated edits.
  • Permissive Content Policy: A more flexible content policy compared to many competing AI video models.