AutoYTPoo

What It Does

• Generates a text description of a strange/humorous image using LLaMA.
• Generates a voiceover script describing that image using LLaMA.
• Enhances the text description and uses Stable Diffusion to generate the image.
• Generates the voiceover audio using Coqui TTS.
• Stitches the image and audio into a video using ffmpeg.
• Generates a title for the video using BLIP.
• Uploads the video to YouTube. WIP

Prerequisites

• Git for Windows
• Conda
• MSVC Build Tools
• ffmpeg (in this dir or on your PATH)

Models

Clone these to the ./models directory in this working copy.

• stable-diffusion-v1-5
• Llama-2-7b-chat-hf
• promptgen-lexart
• blip-image-captioning-large

Random Windows Setup Issues

NOTE After installing MSVC Build Tools, ensure the following components are also installed:

You must also add the Windows 10 SDK path to your PATH environment variable. For example, C:\Program Files (x86)\Windows Kits\10\bin\10.0.19041.0\x64.

I had to run pip install --force-reinstall regex at some point...

Always use Git Bash for terminal commands below.

Create conda environment

conda env create -f environment.yaml

Usage

conda activate autoytpoo

python autoytpoo.py