Work done:
The model was trained on a large dataset.
More than 20 best models took part in the merger
Results:
Improved overall aesthetics and quality
Improved understanding and adherence to prompts
Improved tokens: cinematic, depth of field, semi-realistic