Type | |
Stats | 73 892 327 |
Reviews | (12) |
Published | Feb 9, 2025 |
Base Model | |
Training | Epochs: 15 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | tr4j3 |
Hash | AutoV2 AF69325C02 |
SD15 Version 1
This lora produces bullfighting costumes. It's kind of rough. Use at full strength. Use the keyword tr4j3, and combinations of colors and metal. For example, "salmon pink and gold" or "black and silver" along with "traje de luces". The first version works well, but tends to want to show a behind view. Future versions will correct this. Enjoy!
SD15 Version 3
Much, much better. Reduced dimensions from 128 to 16. Cleaned up captions. Weighted training data so that higher resolution images taken from front got more weight and detailed closeups and views from back were de-weighted. Jacked up learning rate and reduced training resolution because a lot of the training data was lower resolution. Result is a much more stable, crisp, detailed and pleasing image generation. Enjoy!
Pony Version V1
Same data, different model and parameters. Recaptioned everything to get more precise prompting. Next version will add more action and position weighting to make it more flexible.
This model can be used to recreate the whole bullring locale as well as just the outfit depending on the prompting keywords. Doesn't work so well with stock Pony model. I highly suggest using Duchaitens.