Type | |
Stats | 125 15,230 2.5k |
Reviews | (25) |
Published | Dec 7, 2024 |
Base Model | |
Training | Steps: 8,000 Epochs: 40 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | evulchibiart style |
Hash | AutoV2 F5BD135355 |
Updates:
🔹 Uploaded V2 and sample gallery
Details:
🔹 Trigger Word(s): evulchibiart style
🔹 Trained On: PonyDiffusionXL v6 (Base)
🔹 Recommended Strength: 0.8-1.0
🔹 Training Samples Seen: 8000
This is a (IMO) very high quality trained lora on around 270 290 pieces of HD artwork, with three eight runs to find good hyperparameters. It's extremely good at pinup and portraiture of course, but will struggle to have non-humanoids in the images unless you try very hard in the negative prompt.
Versions:
🔹 v1: Self explanatory. Baseline finetuned lora, works best with PonyDiffusionXL-v6.
🔹 v1-NoTE: No Text Encoder version of v1. More flexible for use on SnowPony and other finetunes, style transfers more agnostically.
🔹 v2: Retrained on data with cleaner results, no watermarks on images by default, more epochs.