LegalEagle (RealisticVision LoRA)

LegalEagle v1

Images used to showcase this model were generated using the ADetailer extension.

Model used is Juggernaut (Final).

I strongly recommend using a strength of 0.7-0.8 and a CLIP skip of 1.

In Brief:

This Low Rank Adaptation (LoRA) model is trained on images of the youtuber LegalEagle, and uses RealisticVision v5.1 by SG_161222 as its base model for network weights. This was chosen to achieve better performance in merges that aim for realism while still maintaining SD 1.5 compatibility.

This means using magic keywords like "masterpiece, best quality, high res, ultra HD, portrait" and so on may actually result in lower quality images.

Recommended Settings:

CLIP skip of 1

LoRA Strength: 0.7-0.8

Prompt:

detailed realistic open eyes, {your prompt goes here}

Negative Prompt:

easynegative, bad-hands-5, (anime:1.4), {your negative prompt goes here} 

Samplers:

• DDIM

• DPM2

• DPM++ 3M SDE Karras (recommended)

• DPM++ 3M SDE Exponential (recommended)

The SDE variants of the DPM 3rd gen samplers work really well to address minor overfitting issues.

Recommended Inversions:

easynegative by Havoc

bad-hands-5 by Samael1976

Additional Tools:

ADetailer by Bing-su (extension)