bakamitai-automatic

Generates a Baka Mitai meme from an input image

Google Colab (faster)

1. https://colab.research.google.com/drive/1kUKt-MJ8ThEkqWsiLwRmEJOIdwqwbzyr?usp=sharing
2. Ensure the runtime type is GPU.
3. Press run on each button except the second last one, in order.
   • Wait for previous execution to finish before pressing the next button.
   • Ignore non-fatal warnings or errors. Report fatal errors via the Issues tab.
4. Navigate to the folder named bakamitai-automatic in your Google Drive.
5. Replace input_image.png with desired image. Cropping to face is desired.
6. Press run on the second last button.
7. Answer the Nvidia GPU presence prompt with y.
8. Check the same folder that you navigated to in step 4 for result.mp4.

Local (slower)

Prerequisites

• git, ffmpeg and python3 (version 3.7 is guaranteed to work) installed and added to PATH
• Nvidia GPU (optional)

Running the program

1. Clone the repository.
2. Replace input_image.png with desired image. Cropping to face is desired.
3. Execute start.sh or double click start.bat if on Windows.
4. Wait for packages to install. Torch takes especially long.
5. Answer the Nvidia GPU presence prompt.
6. Wait for result.mp4 to be generated in the directory that start.sh was executed in.
7. Enjoy

Issues

• If Google Colab refuses to cd, factory reset the runtime. Runtime -> Factory reset runtime.

Repetition

To repeat the process, delete generated.mp4 in working-dir (or click the last run button in Colab). Replace input_image.png with another image and

• Colab: press the button that runs start.sh.
• Local: run start.sh or start.bat if on Windows.

Cleanup

• You are free to delete working-dir, but this is equivalent to cloning the repository and setting input_image.png.
• You can delete small files like bakamitai_template.mp4, dmdn.mp3 or generated.mp4.
• It is advisable to keep venv and vox-cpk.pth.tar as they are large and may take more time to redownload.
  • venv may not be present on Google Colab. Ignore this step if so.