For version 2 I went through the dataset and applied photoshop to each image, generally speaking splitting the image into 3 layers, setting the top layer to luminance only, then tuning brightness and opacity, then color and opacity on middle layer to modestly stretch contrast while preserving tone and saturation.

The Noob VPred 1.0 version is not as straightforward of an effect, because the base model is already capable of strong contrast and darker images. So this version helps promote contrast, shadow detail, and lower light levels.

Helps your pictures not be so damn bright. Trained on Illustrious 0.1 for now for broad compatibility. Tested on Noob 1.0, at 1.0 power.

I looked for a lot of dark images with good shadow detail and contrast, less than peak brightness on highlights, and lower average luminance.

To use: load the lora, and use tags like dark, chiaroscuro, moonlight, dark background, black background, etc to shape how bright your picture is. Dark and chiaroscuro are the two big tags, and I specifically included some moonlight pictures for nice highlights at night when you want them.