V2 fix

Trying to correct the Loras I've overtrained, this uses the same trigger word and strength as the V2 from before I just went back and downloaded the lower epoch. Trained on 27 images of Madison.

What to start the prompt with

You need to put "photo of m4d1s0n a woman" at the start. It's what I did while testing and what I found to be best. If you want a half body photo put "half body photo of m4d1s0n a woman". For full body- "full body photo of m4d1s0n a woman" etc.

Strength it works best at

I've tried a variety of strengths and what always works best for me is 0.8. More than that usually messes up hands plus doesn't add to the likeness, at least in my opinion.

You don't need to tag her regular features

You do not need to describe her eye color, hair color, body type, etc. Unless you're in the mood with her having blonde hair or something.

What to put in the negative prompt

ugly, deformed, airbrushed, photoshop, rendered, (multiple people), blurry, child, mouth open

What models this works best with

I haven't had time to test with anything except Amalgamation but these are all the models my Loras usually work well with:

Himeros: https://civarchive.com/models/255907?modelVersionId=423235

Amalgamation: https://civarchive.com/models/287016?modelVersionId=456815

Juggernaut: https://civarchive.com/models/133005?modelVersionId=240840

Psychosis: https://civarchive.com/models/449888?modelVersionId=515453

V2

This new version is trained only on Madison pics from 2020. This version should also be able to work with most models but I like it best with amalgamation and himeros. Recommended strength is 0.8. Sample images are 1024x1024 with Adetailer. I recommend keeping steps at 30-50. Recommended negative prompt: ugly, deformed, airbrushed, photoshop, rendered, (multiple people), blurry, child, mouth open. Hopefully at least a good portion of the sample images pass civits inspection. Idk if it's the model or the lora but this one seems to have a higher chance of revealing clothing/nsfw. If you want sfw images put "nsfw" in negative prompt. May depend on the model as well. This version is not trained with the same captions as v1, I went with less complex/long captions.

V1 (FOR AMALGAMATION)

Checkpoint Recommended/Can it work with other models?

Some of you may remember my previous versions 1 and 2. They were both trained on SDXL base. This was trained on AmalgamationXLv2, you can find it here: https://civarchive.com/models/287016 . I have tested this lora with NewRealityXL and it worked fine, I posted some sample images from it. I still highly recommend using the checkpoint it was trained on. Sample images are generated on 1024x1024 with Adetailer.

Strength Usage

There is no best strength it just depends on what you want. Personally I use 0.7-0.85 a lot. A higher strength usually equals more resemblance to the images used for training. In the sample images I show the result of using a high strength. They are the ones of Madison wearing a blue feathered dress. Using a high strength can sometimes lead to deformity.

Trigger Word

The trigger word is m4d1s0n, you can just add it to your prompt or do "m4d1s0n a woman"

Some captioning help

This is an image followed by its captions. YOU DO NOT NEED TO CAPTION THIS WAY WHEN GENERATING IMAGES THIS IS JUST AN EXAMPLE TO HELP IF ITS NEEDED.

m4d1s0n, The image features a woman looking towards the right, showcasing a fashionable ensemble and a model-like pose. She has press-on nails, bracelets, and rings adding to her accessorized look. The event backdrop includes the ASOS logo, indicating a fashion-related event. The woman is dressed in a black leather crop top paired with ripped blue jeans, exuding a casual yet chic vibe. Her wavy hair is styled elegantly, and she is smiling warmly, wearing hoop earrings and makeup that enhances her features. Standing confidently in high heels, specifically clear stilettos, she exudes a sense of style and confidence. The bright-colored background and wooden floor of the indoor setting provide a trendy backdrop for the photo, emphasizing the contemporary and fashionable nature of the attire and the overall scene.

I have captioned various things that include but are not limited to:

-Hoop earrings, stud earrings

-manicured nails, press-on nails

-bracelets,rings

-wavy hair

-mouth open, teeth visible

-etc

Caution: Using these in your positive prompt will not always lead to exact result. Though at times it may help putting them in the negative prompt.

What to put in negative prompt to avoid some bad hands

In images I trained with hands I usually had to tag "press-on nails, manicured nails, rings, bracelets". So a way to avoid bad hands is to put these in your negative prompt or lower the strength of the lora.