Hey HN — This is a simple web app that allows you to enter text and watch as an in-browser neural network generates the corresponding handwriting in real time. I’m happy to hear feedback or answer questions about how it works!

Very neat! Would you be willing to share more about how you're implementing the in-browser neural network? Doesn't look like webassembly, and I don't see you load any external libraries, just download a binary with content-type "application/macbinary".

I implemented the network from scratch in vanilla javascript. The file is just a custom binary file containing the names, shapes, and weights of the network parameters. If you have more specific questions, feel free to ask them.

That's pretty cool.

Back around 2000 my startup used hand written business cards. There were only a few of us so my artist wife simply wrote them (not using fancy calligraphy, simply clear printed handwriting) and we printed them up on the usual card stock.

I remember handing one to a VC who was mighty impressed and said, "wow, the software you used is amazing. Each character varies slightly. I'm impressed."

We didn't get an investment from their firm though; impressive as the cards may have been our actual product didn't appeal to them.

This was one of graves' cool demos back when he was with hinton. https://www.cs.toronto.edu/~graves/handwriting.html

this is pretty dope. tried it out. though position of "." on top of i's will for sure tell it's robot handwriting.

I love stuff like this! I find it's quite a lot of fun to play with.

Style 6 seems to have a little bug handling the letter combination "ex" at higher legibility. Try the phrase "I had to collect things from my ex" to see what I mean. It goes delightfully crazy off the side of the page.

This is pretty cool!

How does the legibility work? It seems to decrease variation the higher it is.

Outputs are sampled from a probability distribution, and increasing the legibility effectively concentrates probability density around more likely outcomes. So you're correct that it's just altering variation. The general technique is referred to as 'adjusting the temperature of the sampling distribution'.

Really cool I thought it was just you had different datasets labeled for legibility

It seems that the legibility slider is inverted. Low legibility (left most position) is much easier to read than high legibility (right most position) when I tried playing with the slider.

Had a lot of fun playing with it, but the character limit kept me from seeing what a paragraph would look like (spacing between the lines test).

Here's how it looks with multiple lines : https://imgur.com/a/VTpvNS3

Thanks looks great. Might need a bit more spacing between paragraphs but that's really cool.

This is very cool, I'd love to uses this in a personal project, will the js code be on github?

Really neat! Where did you get the training data for? Is there a way to contact you?

The dataset is from the IAM On-Line Handwriting Database. I've linked my github, which has my email, on my profile.

It looks like upper case E rarely has the top line, even on maximum legibility.