@Gargron yeah, it might be better to make that opt-in

@noiob It would be cool to have an actual neural net try to label images. Facebook and Google Photos do this to generate alt text. A lot more involved than OCR though...

@Gargron I've seen it on Instagram and it's really not worth it on there, it's all stuff like "image may contain person, text"

@noiob @Gargron I've experimented with aws recognition but well imagine the wad of bills emoji with wings flying from the community to Jeff bezos

@bonzoesc @noiob @Gargron Plus more privacy-conscious folks will find it creepy.

@mkb @noiob @Gargron literally anyone should find asking a huge dod supplier to look closely at their photos creepy

@Gargron @noiob that is generally a Really Expensive operation and even state of the art ones are somewhat mediocre because you have two operations that are hard in a row, image to concept and concept to text (consider https://twitter.com/picdescbot?lang=en - the descriptions aren't even particularly detailed and they are still wrong or useless most of the time)

@halcy @Gargron yeah, and I don't see people complaining about writing image descriptions, I've seen a lot of folks even get a bit creative with it

@Gargron It can probably be lazy-loaded.

@sorin Hey, tesseract.js tries to load stuff from unpkg by default and I obviously need to self-host everything... Their class takes a script URL for a Worker but I'm not sure how to make it load a local version with webpack? Any ideas?

@Gargron Hey, you probably need to copy the worker script in /static (either manually or via something like copy-webpack-plugin) and then pass that path to the class.