HDR Merge

Exposure fusion, not tone-mapped HDR

This uses Mertens exposure fusion. It weighs every pixel of every frame on contrast, saturation and how well exposed it is, then blends the winners straight into a normal displayable image.

It never builds a radiance map, which is why it needs no EV values, no camera response curve and no tone-mapping pass. It also avoids the halo-and-grunge look tone-mapped HDR is known for: the output reads as a well-exposed photograph rather than as an effect.

The trade-off is that there is no scene-referred HDR file at the end. If you need a 32-bit radiance image for compositing, this is the wrong step.

What your source frames need

Real exposure separation. Three frames around -2, 0 and +2 EV give the fusion something to choose between; three frames at the same exposure give it nothing, and the result will look like the original.

Consistent framing. Fusion needs identically shaped frames, so the exposures are brought to a common size before blending. A tripod or a steady burst produces the cleanest edges; visible drift between handheld frames shows up as softness along contrast boundaries.

A still scene. Anything that moves between frames — people, branches, traffic, water — is averaged across the exposures and ghosts.

The long edge of each exposure is capped at 2400px. Fusing several frames at full capture resolution took long enough to hit the request timeout, so the cap is what keeps the merge returning an image instead of an error.