search

metauni / general

For development of everything else
1 stars 0 forks source link

Image uploading pipeline for speakers #36

Open ecurtiss opened 1 year ago

ecurtiss commented 1 year ago

Currently, if a speaker wants to import images for their talk, they must either

  1. Send them to a metauni developer to manually import
  2. Go to https://create.roblox.com/dashboard/creations?activeTab=Decal, upload their images, and then copy and paste the asset IDs onto image-enabled boards

Option 1 sucks a little, and option 2 sucks a lot. Here is an idea for an end-to-end solution:

  1. Speaker goes to a website and logs in with Roblox oauth
  2. Speaker uploads images to the website, which get uploaded to their Roblox account
  3. Speaker joins a metauni place
  4. Speaker opens a tool that shows them their uploaded images
  5. Speaker clicks on an image and then clicks on the desired board
ecurtiss commented 1 year ago

Input from Dan:

The main reason that image uploading is a pain is (for me) that I have to

  1. Constantly remind speakers of the moderation limits (no URLs, no pictures, no tiny text, no email addresses, ... )
  2. Check every slide to find violations, and manually modify them to fix (by pasting white rectangles, for example)
  3. Upload slides, get my account locked because I miss something, click the "beg for forgiveness" button
  4. Goto (1) until pass

Solution: Run OCR and facial recognition on the uploaded images, and reject an image if any violations are found. Allow a manual override if the speaker is super sure it's safe. If possible, automatically block out/blur the violations and propose the fixed version to the speaker.