datass280/DatAss Rev FINAL - Pony Realism
Information: This model was merged and trained on women only. It can depict males but is not very flexible. LoRA Training: Use either wdv3 captions or real captions, i personally use Florence-2 Large in TagGUI Train on base pony, results are close enough imho Use Prodigy Optimizer with a d-coeff of 0.5 Use masked training with an 1.0 alpha mask for the face and 0.3 for the body, if possible remove hands and feet from the mask Do not train the text-encoders, only unet 600 - 1000 steps is suggested Suggested Settings: Sampler: Euler a + AYS (Align your steps) Resolution: 832x1216 Steps: 14 CFG:
| Author / organization | datass280 |
|---|---|
| Downloads | 8.4K (as reported by Civitai) |
| Likes | 568 |
| Task | text-to-image |
| Created | 10 July 2024 |
| Last updated at source | — |
Silytics shows public metadata from Civitai and links to the original. Information may be incomplete or out of date. Last refreshed 15 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.