Using Teachable Machine

Hiya, I’m trying to see how easy it would be to make a custom OP for Teachable Machines:

Thing I’m kind of running in to from the start is the lack of CDN import. Does anyone have another method for including libraries?



Hi Mick,
Libraries have to be added manually by the undev team.
I’ll put this forward and see if it might be possible to add it in the future.



Hi Andro,

Thanks for the quick reply! Having a simple way to include ML would be amazing, we’re hoping to bring cables back to Hyper Island this year if we can make it fit for the workshops we’ve got in mind :slight_smile:

1 Like

Seconding this but for Three/Aframe/AR.js

1 Like

Third for both ML and Three/Aframe/AR.js!!!

1 Like

And tracking.js :smiley:

(on second thought… maybe a tracking solution that is properly maintained)


we have been looking into ar.js and it is heavily dependent on threejs or aframe doing the rendering. so this will conflict with cables and be shitloads of work to integrate…i really like the library though and we hope to eventually find some time or have the ar.js guys abstract the rendering part out of their library…

on the “teachable machine” thing. i actually do not quite get how it works. is it “just” using this library on your website?

this then would mean we could integrate this library and you then could code ops using it…is that what you are asking for?


@stephan Jonas is part of the AR.js dev crew, he’ll try and nudge them :wink:

On the ML side

instead of teachable machine maybe it would be worth going for ml5js
There are tons of stuff to play with there.

It could also be worth developing examples for runwayml :
There is not much to do library wise since we can use http calls to exchange data or images with it. I already have put an example with the help of stephan for the custom nodes :

but also thumbs up for some ar /xr !

The libraries have been added. This is my first attempt at making an example patch :slight_smile:

Just go go and make your own model


this is amazing! :slight_smile: i tried to play around with your patch (had to copy custom op and add the library)
it works great with your red and silver model, but for some reason i couldnt get it to load my own.

Is it because i trained pose instead of image model?

Otherwise thnk you and great job!

~edit: tried it with a image-based model and it works great!

Yeah I’m still working on the pose example

@kdklv here’s the pose example :slight_smile:


Aaaaand, here’s the whole family

Image Classifier

Pose Classifier

Audio Classifier

The image/pose classifier can only handle input via a 2D canvas element, so you can’t use it with imageCompose. If anyone wants to look at that happy to suggestions.

The audio example doesn’t use the microphone input from Cables, but does it on its own. Nicer would be if you could wire in the microphone OP. Besides that it works.


these made it into cables and con now be used by everyone without using userops or copying code!

1 Like