Modified synthetic dataset based on sd1.5 embedding.
Overall consistency is getting better.
Another test of the CivitAI lora training service.
Synthetic dataset created from an SD15 character embedding I created.
V2 improves consistency.. less artifact and distortion.
