Inspiration

It was a conversation between two members after witnessing the social epidemic that is TikTok and Instagram Reels. Everyone is addicted, including you - the reader - so we need to ensure that same addiction is brought to Man pages. Take in the brainrot, consume it like anything else.

What it does

It turns Linux Man (or pwsh help) pages into entertaining videos to ensure that the listener understands the content and knows exactly what to expect.

How we built it

The main backend was made in Python, with there being 3 main AI components, that being:

  • Prompt generation for scripts and scenes.
  • Text To Speech generation for the video's speaker.
  • Video generation on freeware.

Challenges we ran into

  • Downloading models
  • Download fails
  • Compiling
  • Waiting
  • Updating drivers
  • Reinstalling
  • Rebooting and trying it again
  • Eduroam disconnecting on remote PC (and having to run back to reconnect it)
  • Giving up and wasting 12 hours because AMD ROCm sucks, before fortunately sourcing a newer team member who had a ridiculously expensive GPU that fits the requirements.

We graciously used their PC in halls - working over RDP, so we could use the ComfyUI dashboard as a web UI & API for generating videos. We found Serveo to let us use an SSH to get access to the web portal in A32... some of the time. It crashed frequently when trying to load the web UI, but restarting the tunnel (sometimes up to 10 times) seemed to make it work consistently????

We had lots of problems with mismatching Python library versions. We found a model that would allow TTS with mouth movement - but this would involve having to downgrade our Python version and re-installing everything for the third time. We'd rather not.

Windows Defender silently quarantining files!!!!! (spending hours debugging without realising this)

Accomplishments that we're (sort of) proud of

  • Actual open source video generation of Steve Bagley, mainly spontaneously combusting.
  • Prompt hacking with Gemini to ensure that we get a variety of different scenes and explainers.
  • I got audio working in the last hour! (thought it cuts out a lot between scenes)

What we learned

  • Open source AI video generation is a pain, you might as well shell out for Sora or something.
  • It will probably work better in a couple of years.
  • I now have a justification for buying an RTX 5090.
  • AI won't be replacing Steve's job any time soon.

What's next for Team 15: Mansplain

  • Disbandment. Sorry fans, but this was the golden jewel.

Built With

Share this project:

Updates