Detail Slider LoRA | Illustrious-XL - v1.0 - Initial Release
Highlighted images
Recommended Prompts
1girl, No-LoRA, best quality, absurdres, masterpiece
Recommended Negative Prompts
scenery
painting, blender, lowres, blurry foreground, worst quality, cropped, low quality, missing fingers, extra digit, error, bad anatomy, sketch, normal quality, jpeg artifacts, deformed, disfigured, poorly drawn face, mutation, mutated, ugly, extra limb, poorly drawn hands, missing limb, floating limbs, disconnected limbs, malformed hands, long neck, long body, monochrome, (watermark), signature, logo, name, bar censor, abs
Recommended Parameters
samplers
steps
cfg
resolution
other models
Recommended Hires (High-Resolution) Parameters
upscaler
upscale
denoising strength
Tips
Use input weights between 3 and 4.5 for best detail effects.
Set Hires CFG Scale between 3 and 5.5 to avoid messed up backgrounds, preferably around 3.5 or 4.
Keep sampling CFG Scale values <= 7 for stable image generation.
Adding 'scenery' tag enhances outdoor scenes.
Negative input weights are not consistently effective currently.
Version Highlights
v1.0 Initial Release
👇👀 Read me! 👇👀
This is a slider for IllustriousXL trained on the base model Illustrious XL v0.1. The training data for this slider is the same as my popular detail slider for PonyXL, so there is a little style bleed but nothing you can't overcome with a style LoRA. The goal of this slider is to easily and consistently introduce more detail in generated images, mostly in the background.
No activation tag. This slider LoRA supports bipolar input weights, although negative input weights are not as consistent as I would like in v1.0 . I recommend using weights 3 to 4.5 to add a nice amount of detail to your generated images.
It is important to note that if you're getting weird messed up backgrounds, you should set your Hires. fix "Hires CFG Scale" to anywhere between 3 and 5.5 . Personally I tend to stick to 3.5 or 4. And you should use sampling CFG Scale values <= 7 generally speaking.
Example prompts can be seen in the attached example images.
No special negative prompting needed, although the scenery tag is great to add for any scene that is outside.
Base IllustriousXL does a pretty good job at adding details when you use the correct prompts for them, so this slider LoRA isn't as necessary to use as it is for PonyXL, but it still provides an easy and consistent way to add details to images.
If you generate images with this, please share! I love seeing what people make with my LoRAs and it helps keep me motivated to make more. If you have IllustriousXL LoRA requests, feel free to leave a comment and I'll look into into it 👍
Known Issues:
Negative weight values do not consistently remove image detail
The LoRA somewhat stylizes output images
High positive input weight values (
5>=) make backgrounds fall apart at higher CFG Scale values
Changelog:
December 2, 2024 - Initial release
Model Details
Model type
Base model
Model version
Model hash
Creator
Discussion
Please log in to leave a comment.
