Is it possible to give the neural network and pictogram and train him to generate variant of this picto?

I know there is some initiative to generate vectors drawings like https://github.com/tensorflow/magenta/blob/master/magenta/models/sketch_rnn/README.md or some logos generated by AI but it's not really qualitative…

Any idea/tips?


1 Answer 1


Most classes of generative models of images (with the notable exception of autoregressive models) have a "latent" space.

Typically, a property of a well trained and sufficiently powerful generative model is that traversing the latent space also smoothly traverses the image space in a coherent manner, yielding variants of the starting image.

So in short: train a generative model, then take nearby points in your latent space to obtain "variants"


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.