Memegen

This repository contains code to train text generation models and use them to generate memes. The architectures used are textgenrnn and this

The training code is in the jupyter notebooks meanwhile prediction is done with textgenrnn_predict.py (for textgenrnn) and predict.py (for the convolutional network).

Note that textgenrnn is trained via txt files and not via JSON files like the convolutional networks. The json_to_txt.py script in the scraper folder can be used to perform the conversion.

Disclaimer: the generated memes will often have a poor quality and you may have to generate a large number of them to find "good" ones.

Sample outputs for the two models are in the samples folder

Setup

Run pip install -r requirements.txt

Meme generation (prediction)

The prediction scripts load a trained model, use it to generate text and create a meme using a template from memegen.link. The template is specified via the --template flag for both prediction scripts. Check the link for the available templates!

The prediction scripts output --count images, to be saved in the folder specified by the --out_folder flag.

Convolutional model

To generate new memes using one of the convolutional models run the following command:

python predict.py --model model.h5 --params params.json --count 2 --template drake --out_folder memes

The --boxes flag determines how many text boxes need to be generated for this template.

textgenrnn

To generate new memes using one of the textgenrnn models run the following command

python textgenrnn_predict.py --model model_folder --count 2 --template drake --out_folder memes --boxes 2

If you want to use one of the transfer learning models you need to add the flag --load_only_weights y to the command.

Miscellaneous