Color Palette Analysis

Examination of color distribution including saturation levels, hue diversity, and color balance across the image to detect AI generation artifacts.

AI models are often trained on heavily edited images with boosted colors. They learn to replicate this "enhanced" look, producing colors more vibrant than typical camera output.

Measured using hue histogram entropy, this indicates how many distinct colors are present. AI images may have artificially limited or exaggerated color variety.

By comparing color temperature (red/blue ratio) across image regions. Real scenes have consistent temperature; AI may have unnatural variations.

Yes, color grading can normalize AI colors. This is why color palette has low weight - it's a supporting indicator rather than definitive proof.

Mean saturation above 140 (on 0-255 scale) suggests AI generation. Most camera photos fall in the 40-100 range without heavy editing.

Color palette analysis is not applicable to true B&W images. Other detection methods handle grayscale content more effectively.

Heavily stylized real photos may trigger false positives. The low 3% weight prevents this from significantly affecting overall detection accuracy.

Real photos have varying saturation based on lighting and materials. AI often produces unnaturally uniform saturation across different objects in a scene.

Different AI models have different color biases. Midjourney tends toward vibrant colors, while Stable Diffusion versions vary. The detector looks for general non-natural patterns.