FFMPerative - Chat to Compose Video

FFMPerative is your copilot for video editing workflows. Powered by Large Language Models (LLMs) through an intuitive chat interface, now you can compose video edits in natural language to do things like:

• Change Speed, Resize, Crop, Flip, Reverse Video/GIF
• Speech-to-Text Transcription and Closed-Captions

Just describe your changes like these examples.

Setup

Requirements

• Python 3
• ffmpeg

PyPI:

pip install ffmperative

Or pip install from source:

git clone https://github.com/remyxai/FFMPerative.git
cd FFMPerative && pip install .

ffmperative defaults to inference using the local model on CPU. To use the remotely hosted endpoint, set an environment variable with a huggingface token:

export HF_ACCESS_TOKEN=<your-token-here>

and use the optional --remote flag

Quickstart

Add closed-captions with:

ffmperative do --prompt "merge subtitles 'captions.srt' with video 'video.mp4' calling it 'video_caps.mp4'"

Features

Python Usage

Simply import the library and pass your command as a string to ffmp.

from ffmperative import ffmp

ffmp("sample the 5th frame from '/path/to/video.mp4'", remote=True)

Compose 🎞️

Use the compose call to compose clips into an edited video. Use the optional --prompt flag to guide the composition by text prompt.

ffmperative compose --clips /path/to/video/dir --output /path/to/my_video.mp4 --prompt "Edit the video for social media"

Resources

• ffmpeg-python
• Sample FFMPerative Dataset
• FFMPerative LLaMA2 checkpoint
• Automatically Edit Videos from Google Drive in Colab

Community

• Join us on Discord