ConRad XL WildSwanku

This LoRA was trained on my base SDXL model merge, ConRad XL. It has only been extensively tested with that base model. You are free to try it on other models, but I can't guarantee any results! If you're looking for a more universal Pony XL LoRA in this style, see WildSwanku Pony!

The LoRA does not work as I intend without including the trigger word "wldswnk" in the prompt! I always put it at the end of my prompt as the very last term. Ex. "1girl, maid dress, wldswnk". Use Danbooru tags for best results!

-

The model has a tendency to add signatures or watermarks, often resembling the word "wldswnk". Putting terms like "signature" or "watermark" in the negative prompt can help with this, but I prefer to inpaint/edit them out of images manually if I need them gone.

For settings, I use:

• CFG 3

• Steps 20

• Sampler: DPM++ 2M SDE Karras

Here it is! This is the custom LoRA I trained on my own outputs from the last few years. It's the culmination of my exploration of image generation to discover a style I love to work in. It is definitely tailored toward NSFW outputs. Unlike the base model, I have never encountered censorship in the outputs.

In my experience, outputs with this LoRA upscale very nicely with simple settings on an upscaler such as 4x-AnimeSharp. It should work fine with similar settings on Hi-Res Fix implementations with denoising anywhere from 0.35 to 0.75 depending on how much extra detail/fixes/changes you wish to see. I usually upscale 2x.

I find this model combination is amazing at inpainting with minimal fuss! I have been pleasantly surprised at how I can mask a certain part of the image and describe plainly what I want to see, such as "hand grabbing tail" and it often just works. I typically use inpainting extensively to refine details such as faces, hands, nipples and other fun bits on my favorite pieces. It's not a miracle worker, though. It is much more effective to do a little simple painting yourself to guide the model into what you want.

Please take a look at the showcase on my profile as well as all my image posts to see more examples of what ConRad + WildSwanku can do! I post typical generations (sometimes upscaled) as well as more involved artwork.

-

Please keep in mind, I am not an experienced model merger or LoRA trainer! Therefore, this LoRA may not be very flexible or work at all outside of the use cases I have extensively tested myself.

This LoRA was trained on my own outputs from the following SD 1.5 base models, in various recipes of merges among them:

Wild's Mix

Swoonify

Hassaku

Many thanks to those creators for sharing their work and inspiring me to find my own style!

-

I'd LOVE to see what you can make with my models. Thanks! <3