📚 Join our Skool community for support, premium content and more!

Be part of a growing community and help us create more content like this

Description

An open source automated video creation tool for generating short-form video content. Short Video Maker combines text-to-speech, automatic captions, background videos, and music to create engaging short videos from simple text inputs.

This project is meant to provide a free alternative to heavy GPU-power hungry video generation (and a free alternative to expensive, third-party API calls). It doesn't generate a video from scratch based on an image or an image prompt.

The repository was open-sourced by the AI Agents A-Z Youtube Channel. We encourage you to check out the channel for more AI-related content and tutorials.

The server exposes an MCP and a REST server.

While the MCP server can be used with an AI Agent (like n8n) the REST endpoints provide more flexibility for video generation.

You can find example n8n workflows created with the REST/MCP server in this repository.

TOC

Getting started

• Requirements
• How to run the server
• Web UI
• Tutorial
• Examples

Usage

• Environment variables
• REST API
• Configuration options
• MCP

Info

• Features
• How it works
• Limitations
• Concepts
• Troubleshooting
• Deploying in the cloud
• FAQ
• Dependencies
• Contributing
• License
• Acknowledgements

Tutorial with n8n

Examples

<table> <tr> <td> <video src="https://github.com/user-attachments/assets/1b488e7d-1b40-439d-8767-6ab51dbc0922" width="480" height="270"></video> </td> <td> <video src="https://github.com/user-attachments/assets/bb7ce80f-e6e1-44e5-ba4e-9b13d917f55b" width="270" height="480"></video> </td> <td> </tr> </table>

Features

• Generate complete short videos from text prompts
• Text-to-speech conversion
• Automatic caption generation and styling
• Background video search and selection via Pexels
• Background music with genre/mood selection
• Serve as both REST API and Model Context Protocol (MCP) server

How It Works

Shorts Creator takes simple text inputs and search terms, then:

1. Converts text to speech using Kokoro TTS
2. Generates accurate captions via Whisper
3. Finds relevant background videos from Pexels
4. Composes all elements with Remotion
5. Renders a professional-looking short video with perfectly timed captions

Limitations

• The project only capable generating videos with English voiceover (kokoro-js doesn’t support other languages at the moment)
• The background videos are sourced from Pexels

General Requirements

• internet
• free pexels api key
• ≥ 3 gb free RAM, my recommendation is 4gb RAM
• ≥ 2 vCPU
• ≥ 5gb disc space

Concepts

Scene

Each video is assembled from multiple scenes. These scenes consists of

1. Text: Narration, the text the TTS will read and create captions from.
2. Search terms: The keywords the server should use to find videos from Pexels API. If none can be found, joker terms are being used (nature, globe, space, ocean)

Getting started

Docker (recommended)

There are three docker images, for three different use cases. Generally speaking, most of the time you want to spin up the tiny one.

Tiny

• Uses the tiny.en whisper.cpp model
• Uses the q4 quantized kokoro model
• CONCURRENCY=1 to overcome OOM errors coming from Remotion with limited resources
• VIDEO_CACHE_SIZE_IN_BYTES=2097152000 (2gb) to overcome OOM errors coming from Remotion with limited resources

docker run -it --rm --name short-video-maker -p 3123:3123 -e LOG_LEVEL=debug -e PEXELS_API_KEY= gyor…[redacted]:latest-tiny

Normal

• Uses the base.en whisper.cpp model
• Uses the fp32 kokoro model
• CONCURRENCY=1 to overcome OOM errors coming from Remotion with limited resources
• VIDEO_CACHE_SIZE_IN_BYTES=2097152000 (2gb) to overcome OOM errors coming from Remotion with limited resources

docker run -it --rm --name short-video-maker -p 3123:3123 -e LOG_LEVEL=debug -e PEXELS_API_KEY= gyor…[redacted]:latest

Cuda

If you own an Nvidia GPU and you want use a larger whisper model with GPU acceleration, you can use the CUDA optimised Docker image.

• Uses the medium.en whisper.cpp model (with GPU acceleration)
• Uses fp32 kokoro model
• CONCURRENCY=1 to overcome OOM errors coming from Remotion with limited resources
• VIDEO_CACHE_SIZE_IN_BYTES=2097152000 (2gb) to overcome OOM errors coming from Remotion with limited resources

docker run -it --rm --name short-video-maker -p 3123:3123 -e LOG_LEVEL=debug -e PEXELS_API_KEY= --gpus=all gyoridavid/short-video-maker:latest-cuda

Docker compose

You might use Docker Compose to run n8n or other services, and you want to combine them. Make sure you add the shared network to the service configuration.

version: "3"

services:
  short-video-maker:
    image: gyoridavid/short-video-maker:latest-tiny
    environment:
      - LOG_LEVEL=debug
      - PEXELS_API_KEY=
    ports:
      - "3123:3123"
    volumes:
	    - ./videos:/app/data/videos # expose the generated videos

If you are using the Self-hosted AI starter kit you want to add networks: ['demo'] to the** short-video-maker service so you can reach it with http://short-video-maker:3123 in n8n.

NPM

While Docker is the recommended way to run the project, you can run it with npm or npx. On top of the general requirements, the following are necessary to run the server.

Supported platforms

• Ubuntu ≥ 22.04 (libc 2.5 for Whisper.cpp)
  • Required packages: git wget cmake ffmpeg curl make libsdl2-dev libnss3 libdbus-1-3 libatk1.0-0 libgbm-dev libasound2 libxrandr2 libxkbcommon-dev libxfixes3 libxcomposite1 libxdamage1 libatk-bridge2.0-0 libpango-1.0-0 libcairo2 libcups2
• Mac OS
  • ffmpeg (brew install ffmpeg)
  • node.js (tested on 22+)

Windows is NOT supported at the moment (whisper.cpp installation fails occasionally).

Web UI

@mushitori made a Web UI to generate the videos from your browser.

<table> <tr> <td> <img width="1088" alt="Screenshot 2025-05-12 at 1 45 11 PM" src="https://github.com/user-attachments/assets/2ab64aea-f639-41b0-bd19-2fcf73bb1a3d" /> </td> <td> <img width="1075" alt="Screenshot 2025-05-12 at 1 45 44 PM" src="https://github.com/user-attachments/assets/0ff568fe-ddcb-4dad-ae62-2640290aef1e" /> </td> <td> <img width="1083" alt="Screenshot 2025-05-12 at 1 45 51 PM" src="https://github.com/user-attachments/assets/d3c1c826-3cb3-4313-b17c-605ff612fb63" /> </td> <td> <img width="1070" alt="Screenshot 2025-05-12 at 1 46 42 PM" src="https://github.com/user-attachments/assets/18edb1a0-9fc2-48b3-8896-e919e7dc57ff" /> </td> </tr> </table>

You can load it on http://localhost:3123

Environment variables

🟢 Configuration

| key | description | default | | --------------- | --------------------------------------------------------------- | ------- | | PEXELS_API_KEY | your (free) Pexels API key | | | LOG_LEVEL | pino log level | info | | WHISPER_VERBOSE | whether the output of whisper.cpp should be forwarded to stdout | false | | PORT | the port the server will listen on | 3123 |

⚙️ System configuration

| key | description | default | | ------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------- | | KOKORO_MODEL_PRECISION | The size of the Kokoro model to use. Valid options are fp32, fp16, q8, q4, q4f16 | depends, see the descriptions of the docker images above ^^ | | CONCURRENCY | concurrency refers to how many browser tabs are opened in parallel during a render. Each Chrome tab renders web content and then screenshots it.. Tweaking this value helps with running the project with limited resources. | depends, see the descriptions of the docker images above ^^ | | VIDEO_CACHE_SIZE_IN_BYTES | Cache for <OffthreadVideo> frames in Remotion. Tweaking this value helps with running the project with limited resources. | depends, see the descriptions of the docker images above ^^ |

⚠️ Danger zone

| key | description | default | | ------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- | | WHISPER_MODEL | Which whisper.cpp model to use. Valid options are tiny, tiny.en, base, base.en, small, small.en, medium, medium.en, large-v1, large-v2, large-v3, large-v3-turbo | Depends, see the descriptions of the docker images above. For npm, the default option is medium.en | | DATA_DIR_PATH | the data directory of the project | ~/.ai-agents-az-video-generator with npm, /app/data in the Docker images | | DOCKER | whether the project is running in a Docker container | true for the docker images, otherwise false