Design ads and posters with Machine Learning


I would like to write a program that takes a number of elements as input (some images and some strings) and outputs a design (the program places the images and strings on a canvas, but also styles the strings).

Can someone point me in the right direction of a possible solution to this problem? I've looked for machine learning algorithms for design but have not found a single one to give me any tips on how to build one of these.


Posted 2018-05-15T07:30:12.763

Reputation: 11

You must decide which is your target, quality factor: minimal gaps (total canvas size)? esthetics's ? – pasaba por aqui – 2018-05-15T08:11:16.397

Esthetics's, or at least something that looks human made. But I do not know how to define it. – Natanael – 2018-05-15T08:22:28.080

How would I define the quality measure? – Natanael – 2018-05-16T07:19:32.717

Why is my question down voted? Is there something wrong with it? – Natanael – 2018-05-16T07:23:19.287

Down voted due to too broad and low quality: you can not ask "I want made that, say me how". No pre-work is show. – pasaba por aqui – 2018-05-16T07:26:43.667

But if I don't know how to start even, how can I be more specific? – Natanael – 2018-05-16T07:28:13.923

"intelligent system for page layout": 101 scholar articles, 2.630.000 global results, – pasaba por aqui – 2018-05-16T07:32:35.610

Thank you! That's what I was looking for, some place to start. – Natanael – 2018-05-16T07:38:33.523



If you have images contain styled strings and images contain printed strings, you could use Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks to learn to translate between both distributions where paired training data does not exist (not styled string image, styled string image).

Fadi Bakoura

Posted 2018-05-15T07:30:12.763

Reputation: 348