Per this discussion: https://www.reddit.com/r/MachineLearning/comments/8o846n/d_what_is_the_best_implementation_of_neural_style/

I'm wondering if there has been any updates.

The consensus seems to be that Jcjohnson still reigns king... However, people found ways to speed it up (https://github.com/lengstrom/fast-style-transfer), clean it up (https://github.com/chuanli11/CNNMRF), or combine speed, quality and effectiveness (https://arxiv.org/pdf/1812.05233.pdf).

I see there have been some repos where people pulled or forked to create their own. I'm hesitant to approach these for their lack of comments/support.

I've personally experimented with Jcjohnson and cysmith (https://github.com/cysmith/neural-style-tf) with many happy results. But I'm wondering if people have had luck with others that I've missing.

Ultimately, I'm looking for a method that can provide accurate style transfers with minimum artifacts (time is less of a priority) with video capabilities. I have a rig with 2 x GTX 1070 running ubuntu 16.04 LTS.

Cheers.