(Windows works fine, just a windows specific test is failing for poor reason, working on updating the test!)

Let's be honest, who really knows how to use ffmpeg. Its that tool that is so helpful but not needed enough to justify learning all of its inner workings. The days of scrolling through stackoverflow are over, with llmpeg. This CLI tool interfaces with an OpenAI LLM (and has support for other LLM API providers) to convert your plain text commands into OS/platform specific ffmpeg commands.

llmpeg is a powerful wrapper around the popular multimedia library ffmpeg, designed to simplify video and audio processing for users who are not familiar with the intricacies of ffmpeg commands. By leveraging language model technology, llmpeg allows users to describe their multimedia processing tasks in natural language, and the tool generates and executes the appropriate ffmpeg commands automatically.

• Ease of Use: Simplify multimedia command generation using natural language inputs.
• Powerful: Leverage the full power of ffmpeg without the need to learn its complex options.
• Flexible: Suitable for a wide range of multimedia tasks from simple conversions to complex processing workflows.
• Extensible: LLM Interface allows for other LLM providers (i.e. Amazon Bedrock, Anthropic, etc.) to work as drop-in replacement for OpenAI LLM
• Conversational: Supports back-and-forth conversation with the LLM to fix mistakes or clarify misunderstandings
• Contextual Awareness: llmpeg is provided context on your OS, OS version, ffmpeg version, and ffmpeg executable path to ensure commands are tailored for your platform.

This was developed on a linux system, but should have valid support on both Windows and MacOS. Automated tests and CI are going to be added soon to validate this behavior.

This tool leverages your own LLM API access. In the case of using OpenAI (the default setup), you must ensure that you have your OPENAI_API_KEY environment variable properly set. You also must install openai with pip:

pip install openai

git clone https://github.com/gstrenge/llmpeg.git
cd llmpeg
python3 -m pip install .

Simply run llmpeg "with your instructions in quotes". See examples below for more.

For more detailed information, run llmpeg -h:

usage: llmpeg [-h] [--backend {openai}] [--openai_model OPENAI_MODEL] instructions

Convert your instructions into an ffmpeg command.

positional arguments:
  instructions          A string containing instructions about the desired ffmpeg use

optional arguments:
  -h, --help            show this help message and exit
  --backend {openai}    The backend LLM API provider to use. Defaults to 'openai'.
  --openai_model OPENAI_MODEL
                        The OpenAI LLM Model that you would like to use.

By default, openai_model is set to gpt-3.5-turbo-0125, which as of today, as the strongest model that can be used with a free developer account with $5 of free credit. If you have a developer account and are willing to pay for credits, you can change this to one of the gpt-4 models for better performance and more accurate command generation.

Simple file conversion and downscaling:

llmpeg "convert the file screencapture.webm to screencapture.mp4, but downscale from 1080p to 720p"

Output Command: /usr/bin/ffmpeg -i screencapture.webm -vf scale=1280:720 screencapture.mp4

Wildcards to select multiple files and using shell tools:

llmpeg "convert all mp4 files in the ~/Desktop/filming/sept_22_2023 folder into mov files with the same name, and store them in a new folder called ~/Desktop/filming/sept_22_2023/movs"

Output Command: mkdir -p ~/Desktop/filming/sept_22_2023/movs && for file in ~/Desktop/filming/sept_22_2023/*.mp4; do ffmpeg -i "$file" -c:v copy -c:a copy "~/Desktop/filming/sept_22_2023/movs/$(basename "$file" .mp4).mov"; done

Contributions to llmpeg are welcome! Here's how you can contribute:

• Issues: Submit issues for bugs, enhancements, or features.
• Pull Requests: Submit PRs for bug fixes or enhancements.

llmpeg is licensed under the MIT License. See the LICENSE file for more details.