EtchAI

Classic "Etch-A-Sketch" board transformed with an AI neural style transfer, in which one can make pixelated sketches on the canvas and get an AI stylized image from various artist's style.

DOCUMENTATION IS CURRENTLY BEING WRITTEN TO MAKE INSTALL AND TRY IT AT LARGE

The board supports three possible styles, and two possible modes of applying these styles. The styles can all be combined with the modes to provide six possible variations for sketching.

Hover is a mode-type that sketches the canvas by hovering the mouse over it.

Drag is a mode-type that sketches the canvas by dragging the mouse over it.

Default is a style-type that colors the canvas with solid black.

Transparent is a style-type that colors the canvas with low-opacity solid black, which gets progressively darker as each pixel area is passed over repeatedly.

Colorized is a style-type that colors the canvas with a random solid color for every pixel area.

render is used to render the current sketch using the selected model.

The Clear button empties the canvas of any applied changes, but retains the selected mode- and style-types.

menu Option such as Clear and Set to auto-refresh.

The Resize button creates a prompt that accepts a given value from 1 to 60, which will change the canvas to represent that value for every row and column (for example, passing a value of 20 will create a grid of 400 pixel areas).

Finally, the label Etch-A-Sketch also acts as a reset button. Pressing it will remove all applied changes, including on the canvas and the selected mode- and style-types.

That's it!

---

Forked Intentions

• Etching and getting a stylized by AI interpretation on the side.
• Peak on what neural network will offers to us

---

About the UI

A Javascript web Ui of Ben Smith's Etch-a-sketch has been hooked up with code to generate an AI stylized image on the side.

Versions

When running the web server, you will have various version you can use.

see: index.html index.html.old index2.html v1.html v2.html v3.html

ie: http://127.0.0.1:8080/v1.html

Docker server

docker run -it --rm -p 8080:8080 docker.io/guillaumeai/ast:etchai

Local server

npm i http-server
cd $MYREPO
http-server

limitations and Issues

• The process of rendering AI on the side is really consuming resources. In the process of making that a bit more effective, You will notice UI glitches, button not responding. Just Click the "Hover" button on the top or the "Size" button
• Hey, get involved, debug it :) :)

Astia Service

That's the service you are require to run on the side for etching. (note that some model you can choose on the top of the right (AI) are personal model that are not public. Bellow are some of the public models I packed in a Docker container your can run.)

Run an Astia Service on Docker

The trained AI model is packed in a Docker repo docker.io/guillaumeai/ast

# docker Launch

modeltag=picasso
serverport=9010
docker pull guillaumeai/ast:$modeltag
docker run -it --rm -e SPORT=$serverport -p $serverport:8000 guillaumeai/ast:$modeltag

docker pull guillaumeai/ast:cezanne
docker run -it --rm -e SPORT=9007 -p 9007:8000 guillaumeai/ast:cezanne

# Other model (see in the index.html)
# TODO Write a Running Script
el-greco
9041

morisot
9044

gauguin
9042

kirchner
9043

monet
9014

munch
9045

peploe
9046

pollock
9047

roerich
9048

van-gogh
9011