Blend modes explained
A blend mode is a rule for combining two pixels into one. Each rule is a line of arithmetic, and once you have seen the line you know what the mode will do to any pair of photos before you try it.
Two pixels in, one pixel out
When one picture sits on top of another, every point of the result is worked out from two values: the base (the picture underneath) and the top. The arithmetic runs on each colour channel separately — red, green and blue — with 0 meaning black and 1 meaning full brightness. Normal simply shows the top; the other modes mix the two in different ways, and opacity then decides how much of that mix replaces the base.
The formulas below are the ones in the W3C Compositing and Blending specification that browsers and design tools follow. On this site they are computed in plain JavaScript rather than left to the browser’s drawing engine, so the double exposure and photo on photo tools save the same file in every browser.
Every mode, with its formula and two worked pixels
The two right-hand columns are computed by the same function the tools use: a dark base under a mid top (0.20 and 0.60), and a light base under a darker top (0.70 and 0.40).
| Mode | Formula | 0.20 under 0.60 | 0.70 under 0.40 |
|---|---|---|---|
| Normal | result = top | 0.60 | 0.40 |
| Screen | result = base + top − base × top | 0.68 | 0.82 |
| Lighten | result = max(base, top) | 0.60 | 0.70 |
| Multiply | result = base × top | 0.12 | 0.28 |
| Overlay | base ≤ 0.5 ? 2 × base × top : 1 − 2 × (1 − base) × (1 − top) | 0.24 | 0.64 |
| Soft light | top ≤ 0.5 ? base − (1 − 2 × top) × base × (1 − base) : base + (2 × top − 1) × (D(base) − base) | 0.25 | 0.66 |
| Difference | result = |base − top| | 0.40 | 0.30 |
In soft light, D(base) is √base when the base is above 0.25 and ((16 × base − 12) × base + 4) × base below it — a curve that keeps very dark values from being pushed around too much.
Which to choose
Screen can only lighten. Black in the top picture disappears and white stays white, which is why a portrait on a white background screened with a forest gives the classic double exposure: the trees show through the dark parts of the person and the background stays clean. Lighten is a harder version — whichever value is brighter wins outright, channel by channel, so edges are crisper and colours can look patchy.
Multiply can only darken, like two slides held up to the light together. White in the top picture disappears, so it is the mode for laying a signature, a pencil drawing or a scanned stamp onto a photo without its white paper. Overlay multiplies in the base’s shadows and screens in its highlights, adding contrast and colour from the top picture — strong, often too strong at full opacity. Soft light is the gentle version: the top picture tints the base rather than replacing it, the usual choice for a texture or a colour wash.
Difference is less an effect than a measuring tool: identical areas go black and the brighter a point, the more the two pictures disagree there. It is what the compare photos tool’s difference map shows, amplified so small changes are visible.
Opacity and masks come after the mode
The mode decides what the mixed pixel would be; opacity and any mask decide how much of it is used. At 60 % opacity, screen with a base of 0.20 and a top of 0.60 gives 0.68 from the formula, and the saved pixel is 0.20 + 0.6 × (0.68 − 0.20) = 0.49. A luminance mask does the same per pixel, using the base’s own brightness as the opacity — which is how the double-exposure tool confines a texture to the bright or the dark parts of a portrait.
Blend modes act on colour values only. None of them knows where a person or an object is; for that the double-exposure tool offers a mask, and the AI photo merge tool cuts a person out with an on-device model.