With the same training method employed, I have utilized a different model that has yielded superior results. Notably, this new model exhibits reduced loss of detail during the generation process. It excels particularly in generating realistic faces and hands, showcasing a noticeable improvement over the previous model. Additionally, the variety of outputs has significantly increased.
Furthermore, the color saturation issue observed in the previous model has been rectified in this new iteration. The generated images now possess a more balanced color representation, avoiding the excessive saturation of the previous version.
While using this model, you may notice that the generated faces have a distinct "Anime" style rather than a Western one. If the Anime-style faces are not to your liking, I recommend utilizing Version 3.0 temporarily until I can refine the new model further. The primary difference lies in the appearance of the eyes, as they tend to resemble Anime eyes. However, generating images with black eyes should result in eyes that are closer to the Western style.
