NeutralCapture is a small experimental LoRA designed to slightly stabilize lighting balance and midtone contrast in otherwise flat images. It was originally trained during testing of several dataset normalization methods and ended up producing a subtle “neutralized” look that removes a lot of stylistic variance.

The LoRA does not introduce any particular style. Instead it tends to push generations toward evenly lit, plain compositions with reduced color vibrancy and minimal texture exaggeration. In many cases the effect is barely noticeable unless compared directly against baseline generations.

Because of this, it is mostly useful for testing prompt behavior or observing how other LoRAs interact when stylistic influence is intentionally minimized.

This model was not designed for aesthetic enhancement and will usually make images appear more ordinary or neutral

Recommended Settings (experimental):

• Weight: 0.18 to 0.42

• CFG: 4.5 to 6.5

• Sampler: Euler a or DPM++ 2M

• Steps: 18 to 24

• Resolution: 768 or lower recommended

Notes:

• Works best with simple prompts and neutral language.

• Tends to suppress strong stylization LoRAs.

• May produce flatter lighting and lower color saturation than expected.

• Mostly intended for testing prompt interactions rather than artistic output.