Type | |
Stats | 158 415 |
Reviews | (29) |
Published | Jun 7, 2024 |
Base Model | |
Training | Steps: 2,100 Epochs: 20 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Hash | AutoV2 F6634DB990 |
The long awaited CivitAI bikini lora (for pony), this model was created for TheAlly simply because it was mentioned in office hours, not because it was a request. After taking a peek at the 1.5 model and seeing that it seemed to have been made only using the product images I decided to manually edit images to make a larger working dataset which seemed to have paid off.
Use
This model was trained only on one piece, and bikini swimsuits so that will be where it shines, that being said when using "patterned" and sometimes strengthening using the color blue or increasing model strength or triggers you can put the pattern on many things.
This model was trained on pony base, but it can work with xl (the semireal x/y plots are done with voight).
the full positive prompt for x/y examples is listed above the images(i use comfy and metadata makes me cry so I'm sorry i cant attach prompts to images)
Triggers
civitkini - the caption made for bikinis though just using bikini without this trigger seems to work just fine
civitonepiece - the caption for one piece swimsuit same as above it isnt entirely necessary for the effect
these triggers seem to be more useful as a way to strengthen the pattern to be applied to things that are not swimsuits or clothing '((civikini, civitonepiece)), patterned' is good to add to a prompt when you want the bikini pattern on something that the model isn't trained on.
as always, have fun and experiment, lower strength will make a nicer image typically but it will reduce the quality of the texture so there is a trade off, longer prompts and 1.0 strength can help you get the best of both.