Training for this model was conducted locally utilizing SD-scripts.
A minimum configuration of 30 sampling steps is recommended. If generations appear unstable, ensure you are utilizing the appropriate base model rather than a derivative merge. Please message me for questions or concerns.
DISCLAIMER
This model is restricted from commercial use or any generation workflows that monetize the underlying dataset to respect the original creators of the dataset. The creator of this model holds no responsibility for user-generated content. Users are expected to utilize this tool responsibly.
Description
Same training dataset, network/conv dim/alpha were simply adjusted, halving the file size. It behaves much better with upscaling and has higher fidelity to the actual style.