What each measurement means
The analyzer describes your thumbnail with measurable properties. They're visual characteristics that may affect readability at small sizes — not a score and not a prediction of click-through rate.
- Brightness — the average perceived lightness of all pixels (Rec. 709 luma), from 0 (black) to 100 (white).
- Contrast — RMS contrast: how spread out pixel brightness is. Low contrast images can turn into a flat mid-tone when shrunk. The tonal range shows the gap between the darkest and lightest 5%.
- Saturation & colorfulness — average color intensity, plus the Hasler–Süsstrunk colorfulness metric widely used in image research.
- Color diversity — how many distinct colors are used and how evenly. A handful of strong colors usually survives downscaling better than many similar ones.
- Text-like area — regions with dense, high-contrast strokes arranged horizontally. It's a heuristic, not text recognition.
- Whitespace — the share of the frame that is flat and low in detail, giving the eye room.
- Edge density & complexity — how much of the image sits on a strong edge, and the spatial information (SI) measure from ITU-T P.910.
- Subject position — where the most distinctive detail and color concentrate, mapped to a rule-of-thirds grid.
How it works
Your image is decoded by the browser and drawn to an off-screen canvas at 640 px wide. All analysis runs on those pixels in JavaScript on your device. Nothing is uploaded.
- Luma, saturation and a color histogram are computed for every pixel.
- Dominant colors come from deterministic k-means clustering of a 6,000-pixel sample.
- A Sobel filter measures edges; the frame is divided into blocks to classify detail and text-like areas.
- Face detection uses your browser's built-in Shape Detection API when it's available. We don't download a model or send the image anywhere.
Example: reading the results
Suppose a thumbnail shows brightness 34, RMS contrast 9%, and a text-like area of 18%. The low contrast means text and background have similar brightness — even if their colors differ. Open the Readability Tester and check whether the words are still legible at 25% and 15%. If they aren't, increasing the brightness difference between text and background is a measurable change to test.
Limitations
- These measurements don't predict click-through rate, views or how YouTube recommends videos.
- Text-like detection can flag logos, fine patterns or foliage, and can miss very large or low-contrast text.
- Subject position is based on detail and color distinctiveness, not on recognizing people or objects.
- Face detection is only available in browsers that ship the Shape Detection API.
- Upload limits and recommended sizes are set by YouTube and can change; check YouTube Help for current values.