torphix / ai_art

Product Specification

Frontend

• Landing page
• Art gallery:
  • Visually customisable
  • Invite others to view it via email
  • Can trade with other users
  • Can set up auctions to sell your artwork
  • 3D navigatbale space using three.js
• Art generator:
  • Connects to a websocket workspace that lets you:
    • Generate images from scratch with prompt assits
    • Edit images using instruct-pix2pix
    • Add images to your gallery
    • Mint your artwork as an NFT
• Augmented reality app:
  • Pin your artwork at locations around the world where others who use the AR app and look at the same thing can see the artwork you've pinned up

Backend

• S3 bucket to store the generated artwork
• Websocket server that manages the art generator sessions (built on forked)
• A way to easily add models to the backend websocket manager
• Ability to convert artwork into NFTS
• Abiltiy to purchase the art generation tokens using fiat (stripe) or crypto

Deep learning

• Build some of the models yourself eg: style transfer and train in the cloud
• Fine tune instruct-pix2pix some more (extra marks)
• Find ways to gather data

TODO:

• Get the websocket backend working:

  • User can create a session booting up a model
  • User can inference the various models

• Frontend:

  • Gallery:
    • Switch from loop based animation to event based update ie: should only rerender frames when user walk around ie: not every second

Gallery setup and design:

- Uses a list called updateables that hold objects with a tick function that is called every frame update (note only objects that need updating require this)