videopython

Minimal, LLM-friendly Python library for programmatic video editing, processing, and AI workflows.

Get Started API Reference

from videopython.editing import VideoEdit

edit = VideoEdit.from_dict({
    "segments": [
        {"source": "intro.mp4", "start": 0, "end": 3,
         "operations": [{"op": "resize", "width": 1080, "height": 1920}]},
        {"source": "raw.mp4", "start": 10, "end": 25,
         "operations": [
             {"op": "resize", "width": 1080, "height": 1920},
             {"op": "resample_fps", "fps": 30},
             {"op": "fade", "mode": "in", "duration": 0.5},
         ]},
    ],
})
edit.run_to_file("output.mp4")

run_to_file() streams ffmpeg decode → per-frame effects → encode, so memory stays bounded (~O(1)) even for hour-long sources — no frames are held in RAM.

Core Editing

Cut, resize, crop, change speed, freeze frames, silence removal. Multi-segment editing plans with automatic fps/resolution matching for concatenation.

Effects & Audio

Blur, zoom, color grading, vignette, Ken Burns, text and image overlay. Load, normalize, time-stretch, and mix audio tracks.

LLM Plan Schema

Drive videopython from your own LLM: JSON Schema generation, dry-run validation, and a structured repair / normalize refine loop. Learn more →

Automatic Editing

Hand AutoEditor your clips and a one-line brief — a local vision LLM selects and orders the shots, then renders the cut. Learn more →

MCP Server

Expose the auto-edit pipeline as Model Context Protocol tools, so an agent like Claude drives editing with its own model — analyze, browse scenes by keyframe, author a plan, validate, render. Learn more →

AI Generation

Generate images, video, speech, and music from text prompts. SDXL, CogVideoX, Chatterbox Multilingual TTS, MusicGen — all local, no API keys. Learn more →

AI Video Analysis

Transcribe with speaker diarization, classify ambient audio, detect scene boundaries, and describe scenes with a vision-language model. One VideoAnalyzer runs the full pipeline and returns a serializable VideoAnalysis. Learn more →

AI Dubbing

Translate speech, clone the original voice, and re-time the dub onto the source — all in one pipeline. Whisper + a local Ollama model + Chatterbox + Demucs. Source-prosody-conditioned expressiveness and a transcript-quality gate that rejects garbage input before paying for translation and TTS. Learn more →

Installation

pip install videopython              # core editing
pip install "videopython[ai]"        # + ALL local AI features (GPU recommended)
pip install "videopython[ai,mcp]"    # + MCP server (videopython-mcp)

[ai] is the single AI extra and installs every AI capability; heavy ML deps still load lazily at first use. [mcp] adds the MCP server.

Python >=3.11, <3.14. AI features run locally -- no cloud API keys required.

See the Installation Guide for FFmpeg setup and details.