Color coded, masked training

[tncrdn]

Hello. With Onetrainer masked training I realized that if you could take it one step further and mask individual elements in an image with individual color codes and caption them targeting these color codes, you could get much better results from a text to image model. Like in this example someone posted for instance https://www.reddit.com/media?url=https%3A%2F%2Fi.redd.it%2Fhohmjjdyj1ic1.png I was wondering what you think about it and if this or any other method could be implemented in the future for more precise training.