Join our Discord for more images and upcoming updates!

Real - Lily v5 XL

SDXL version of v5. Same training images (already in XL size).

Model: Big Lust v1.0 (SDXL)

SDXL ≠ Pony You should use SDXL models for generation. Mixing this LoRA with pony models will not give you the same quality.

The LoRA works best for generating of close up's or inpainting. On body images you need to inpaint the to get the desired look.

Training time: 10h (RTX 4090)

Training Resolution

• Resolution 1: 832x1248, 2:3, Portrait

• Resolution 2: 1024x1024, 1:1, Square

• Resolution 3: 1248x832, 1:1, Landscape

For tags see the description of v5.

Real - Lily v5

V5 was trained on 257 images from in close up from front and rear position including toys and . All images are hand picked, cropped and improved manually.

Training Resolution

• 627x941 (2:3) Portrait

• 768x768 (1:1) Square

• 941x627 (3:2) Landscape

Model: Lazymix Real Amateur 3.0a

Preview images are generated with 2.25x HighRes fix with minor inpainting

How to use this LoRA?

Follow the link to see tagging details and how to mix this LoRA with my other LoRAs.

Example prompts

_lily_toys, close up photo of a woman  her  with a blue  from below, front view,  with tip inside ,  pulling out, normal ,  having medium outer  and short inner , wet , shaved 

_lily_toys, photo of a woman with blond hair and small   her  with a  from below, front view,   wearing ,  half inside ,  pulling out, normal ,  having short outer  and short inner , wet , shaved 

_lily_front, close up photo of a woman showing her , front view, normal ,  having short outer  and short inner , wet , shaved , , lying in bed

_lily_rear, extreme close up photo of a woman showing her , rear view, normal ,  having short outer  and tiny inner , dry , shaved , 

_lily_front, extreme close up photo of a woman showing her , front view, normal ,  having medium outer  and short inner , peehole, tiny  gape with pink channel, creamy, shaved 

Have a look at the promo images for more examples.

Older versions

v4

Introduction

This is my first release after long testing with LoRA generation. I hope you like her.

I am planing to release other LoRA with similar tags in the future so you can mix different shapes and create new styles. I have already another LoRA in beta phase where this works quite good.

Training Data

This LoRA was trained with close up images of the same to get consistent images. It is capable of generating dry , wet , spread and also some creamy versions. Front and rear view images are used in this version.

Image count: 83+25 (v4)

Resolution: 512x768

Model

v3: LazyMix+ (Real Amateur ) v2

v4: LazyMix+ (Real Amateur ) v3.0a

For inpainting use the same model but converted it to an inpaint version with the checkpoint merger of webui. Inpainting this LoRA without an inpainting model will be hard.

Don't forget to switch back to the regular model if you get weird results. Happens to me all the time when I start a new image. The inpaining model can not produce normal images.

Tags

_lily_front, _lily_rear - Main tags, Only use one

posing - Normal

spreading - will be more open and/or spread with fingers

finger nails - Can be used to add fingers to the image or them (negative)

- Changes poses so buttholes are better visible

wet - Aroused with wet skin

creamy - White cream inside

photo - Training image was a real photo

rendering - Training image was generated with an earlier LoRA version (experimental)

How to use

DPM++SDE Karras: More realistic images

Euler a: Smoother images

Use 512x768 for image size and upscale later.

CFG scale depends on the tags you are using. Anywhere in the range from 2 to 5 should work. Lower values are better.

Try to avoid Restore Faces if possible as it will sometimes changes the to some kind of weird face or adds teeth.

For basic close up images just add the LoRA + _lily_front tag. For more tags see the LoRA metadata. If you add too much tags the result may get worse.

For body shots you need to lower the LoRA strenght but this will reduce details. I try to use the Regional Prompter extension so I can keep the strenght higher. Fix anything with inpainting that you don't like. Or just create an image without the LoRA and inpaint the with LoRA afterwards.

Also works good to inpaint on real HD images to replace a . This was the primary purpose why I created this LoRA.

Upscaling

For best results I am using ControlNet with type tile + Ultimate SD upscale to keep the image stable without adding ghost all over the place. Upscale 2-3x works most of the time.