Concordant - General Negative Embedding版本Concordant-2000 (ID: 115609)

Concordant is another TE negative I made for myself as an experiment but also again figured others might like it's effects.

It won't increase colour or make things more colourful, but instead average out colours without making the image washed out (so, no hyper contrast colours but also less drab greys unless intentional).

It also does add detail, but in a more neutral way that doesn't override style. The training data has 0 people/animals/direct subjects and instead is based entirely on abstract concepts to push images away from concordant layouts and colours (hence the name). The effects work better on semi-real models, but tests were done over multiple models.

2 types, depending on the training steps, as they both give slightly different results. The 5k one also is likely to generate backgrounds instead of blank/plain/simple ones and that helps me a lot as rarely do I want just a blank colour.

All of the tests are without VAE so some washed out images are due to the models wanting VAE. Also no post processing or fixes.

描述:

2000 step version

训练词语: Concordant