This LoRA is designed to push Krea 2 toward more convincing amateur photography, natural skin texture, and less polished image generation.
It is a concept slider rather than a traditional style LoRA. It does not teach the model a new visual style. Instead, it reinforces details and photographic characteristics that are already present in Krea 2's learned knowledge.
Recommended Strength
Recommended range:
1.0 to 2.0
My current preferred strength is:
1.5
This is a very strong LoRA, so values above 2.0 may produce exaggerated results.
Color Correction
The raw output can be heavily oversaturated.
This is a side effect of the positive and negative concept prompts used during training. Removing the relevant color-related concepts significantly weakened the LoRA.
For best results, use a color correction node after VAE Decode.
My current setup uses the official ComfyUI Color Correct node with saturation set between:
-15 and -20
You may also need to reduce contrast, temperature, or other values depending on the image.
The example images have been color corrected. Do not expect the raw LoRA output to match them without a similar correction step.
Sampler Setup
I currently use a two-pass ClownSharkSampler workflow.
First Pass
Sampler: euler
Scheduler: beta
Steps: 12
Second Pass
Sampler: res4s_munthe-kass
Scheduler: kl_optimal
Steps: 3
My workflow is almost identical to the setup explained in this post:
