Attempting to push stable yogi's realism illustrious and the WAI Illustrious models towards quality and realism via embeddings with high quality photos (using sfw DIV2K public photo dataset, captioned with tags combined with brief caption phrases).

Promising results, training slow on 384 images. Will be uploading different strengths.

I can tell you that the example images are from kohya, and while they do seem like nightmare fuel, the embeddings are still pretty functional across the strength spectrum. I'll try to run some better prompts through once the training is complete...