π¬ text2youtube
Description
text2youtube uses cutting-edge AI power for YouTube automation, transforming simple text into engaging video content. Designed with YouTube automation and text-to-video technology in mind, this project streamlines the process of video content creation. It employs advanced AI for scenario generation, text-to-speech synthesis, and video compilation, making it an innovative tool in the field of text-to-video and YouTube automation.
Features
π Automated video script generation from prompt + reference information (like from another video script). We not only generate the script itself but also queries to Youtube/Storyblocks so that we can compose our video later from a set of clips we download from those queries.
π Video voice generation with Bark. Bark is the best for naturally sounding voice at the moment, voice is one of the most important parts of a Youtube video and we did a bunch of experiments there. It runs quick on Google Colab with A100 GPU attached.
π¨ We have some basic code for stitching videos together with MoviePY but it works real slow, so for now we just export the clips from Storyblocks/Youtube and voice-over and stitch them together in Adobe Premier which takes seconds not minutes.
Demos
Getting Started
- Install the required dependencies by running
pip install -r requirements.txt. - Install text-to-speech
pip install git+https://github.com/suno-ai/bark.git - Ensure the
cookies.jsonfile is present with necessary credentials for storyblocks.com website access. - Ensure
env.yamlfile is present providing OpenAi API key and working directories - Provide the necessary OpenAI prompt in prompts directory
- Provide the necessary content inputs in text files in SOURCE_DIR
- Run the main application file
app.pyto start the content creation process. - The program will handle the rest, creating engaging videos with captivating voiceovers.
Project Structure
The project is organized into several modules:
src.audio: Contains audio-related functions for text-to-speech synthesis.src.config: Stores configuration settings for the program.src.logger: Implements logging functionality for the application.src.openai_generation: Handles interactions with OpenAI's ChatGPT for scenario generation.src.video_processing: Manages video downloads from YouTube or videoblocks.com.src.utils: Contains utility functions for data processing and file handling.src.video: Includes video-related functions for compilation and editing.
Results
While developing, we decided to focus on the finance niche and create a youtube channel on Economics/Countries topic. Here is the channel
We posted 1 video daily for about 20-25 days. In the end, we got around 8,000 views, 221 watch hours, and +70 subscribers, average watch time of the video was around 30%. The results of the experiment are pretty encouraging because the videos are not of very high quality and I expected worse.
What To Improve
- It's hard to change the quality of naturally sounding voice generation at the moment but I think we can have a big boost in video generation itself. For example, finding a way to make high-quality text overlays on top of video clips would give a big advantage. Or parsing images from google and applying parallax (ken burns effect) on top
- Improving video generation speed with MoviePY, so videos wouldn't have to be exported into adobe premier pro.
- Allow option to create youtube shorts content, though there are plenty projects which already do that.
Dependencies
- OpenAI GPT-3 API: For scenario generation using language models.
- Neural Network Text-to-Speech Library BARK: For realistic audio synthesis.
- Python Requests Library: For website access and video downloads.
- MoviePy Library: For video compilation and editing.
Contributors
License
This project is licensed under the MIT License. Feel free to use and modify it according to your needs.
