Questions about datasets
Hello! I am very optimistic about the future prospects of this model. Increasing the parameter count to 2.9b undoubtedly raises the upper limit of the model. However, I have some questions about the training set:
Will this model be added to high-level aesthetic datasets such as ArtStation in the future? On sdxl, chenkin noob used the full dan+artstation dataset, which gave the model an unparalleled aesthetic level compared to other sdxl models (making it my favorite sdxl model). Anima's textencoder and VAE have great advantages over sdxl, but to be honest, its aesthetics are somewhat sad compared to chenkin noob
In addition, ArtStation's dataset seems to have greatly enhanced the metallic texture, conceptual understanding ability, and light and shadow effects of Chenkin Noob
Certainly