Today it is common to see how companies embrace the artificial intelligence to improve their products and services. Its implementation is intended to simplify tasks and make your software proposals more efficient, however, some curious people are using it for mere experiment or fun. Dylan Wenzlau, founder of the Imgflip platform, was given the task of creating a meme generator thanks to machine learning, and the result has been very interesting.
Its tool allows you to choose between 48 images Quite popular for making memes, from shocked Pikachu to Philip J. Fry staring in disbelief. The text is automatically added, and while sometimes their message doesn’t make any sense, they get to be funny enough to steal your smile. Others, on the other hand, are very well achieved, so much so that appear to have been created by a human. These are some of the memes we got; you can try it yourself in the following link.
Wenzlau, through a Medium post, explained how he created “This Meme Does Not Exist”. First of all, your software is supported by convolutional neural network, commonly used in automatic image, video and text processing. These networks are capable of identifying the characteristics of any image and acting accordingly. The developer used the technologies of Tensorflow —From Google— and Keras to make it possible.
Of course, any neural network needs material to train and offer better results. Wenzlau already had a platform to create memes manually, imgflip, in such a way that he already had a data gold mine created by his community. Machine learning allowed the tool to be able to identify the basic context of memes according to the background image. The next thing was to generate text and complete the image.
Although you have the possibility to enter some words, even names, the software is not always able to adapt your suggestions to the meme. It is an experiment that is still in the development phase, so it still needs to evolve. Wenzlau has released all the code of “This Meme Does Not Exist” in Github so that others can analyze or improve it. It seems that Skynet is close to having a funny facet.
👇 More in Hypertextual