
Making Enfuse's exposure preference very narrow does not guarantee that every out-of-range contribution disappears. In Enfuse 4.2, if all candidates at a pixel position receive zero weights, the documented fallback assigns equal weights, 1/N. With the finite-support exposure functions, a region you meant to reject can therefore become an average of the available inputs.
Recognize a weighting fallback, not recovered detail
Imagine an exposure series in which every source records a particular bright region outside your chosen acceptable band. Changing the weighting curve cannot create a correctly exposed sample that was never captured.
An averaged result in that region is not evidence that Enfuse found a hidden good exposure. It can instead be the normalization fallback described in section 5.2.1 of the official 4.2 package manual. This guide concerns that specific condition, rather than every possible cause of an unwanted contribution.
Distinguish the function families
| Exposure weighting function | Behavior away from its optimum |
|---|---|
| Gaussian, Lorentzian | The manual describes tails that do not become exactly zero |
| Half-Sine, Full-Sine, Bi-Square | Can assign exactly zero outside their supported interval |
Select the function with --exposure-weight-function. --exposure-optimum chooses its preferred luminance; --exposure-width controls the accepted width. For the finite-support functions, narrowing the width makes the all-zero condition a possibility when no input supplies a candidate inside the band.
The manual normalizes pixel luminance independently of input integer bit depth. Do not use an 8-bit code value such as 128 as though it were the normalized optimum of 0.5.
Inspect the candidates before tightening the band
At the suspect region, compare every recorded source. Ask whether one actually falls in the intended exposure interval and whether that candidate remains eligible. Then review the function and width together.
The manual identifies two ways to avoid this particular pathology: choose a sufficiently broad width that weights do not vanish, or supply inputs spanning enough luminance at every pixel. It also documents --exposure-cutoff as a separate way to set lower and upper cutoffs. Do not treat that option as evidence that an absent good source has been recovered.
The HDR moving-person guide addresses a different problem: sources disagree because the subject moved. Broadening exposure weights does not align that motion.
Keep rejection and missing coverage separate
Input alpha is another mechanism. Enfuse treats a zero input-mask value as exclusion and nonzero values as eligibility. If all input masks are black at one position, the manual says the output has a hole. That is different from eligible candidates whose computed weights all become zero and are averaged by fallback.
For a proposed diagnostic, compare the source values, input eligibility and the chosen weighting curve before judging the output. Preserve the distinction between fusion and display capability too: the HDR gain-map guide concerns a separate representation issue. No exposure fusion or curve comparison was performed here.
Related tasks
- Why Does a Person Look Transparent After an HDR Photo Merge?
- Why Does My HDR Photo Lose Its Bright Highlights After Editing?
Sources and scope
- Enfuse 4.2 official package manual: Enfuse 4.2 section 5.2.1: Gaussian/Lorentz nonzero tails versus finite-support Half-Sine/Full-Sine/Bi-Square; all-zero weights fallback equally 1/N; width/source coverage remedies and separate exposure-cutoff. Section 7.1 separately establishes all-black input-mask holes.
Research checked on 2026-10-23 (Asia/Shanghai). The official SourceForge archive returned HTTP 200 and its bundled 4.2 HTML manual was extracted and read locally; this is not a claim that the public manual HTML endpoint was readable. Enfuse 4.2 section 5.2.1: Gaussian/Lorentz nonzero tails versus finite-support Half-Sine/Full-Sine/Bi-Square; all-zero weights fallback equally 1/N; width/source coverage remedies and separate exposure-cutoff. Section 7.1 separately establishes all-black input-mask holes. Commands and comparison procedures are illustrative and were not executed; the cover is decorative.
Prepared by Imagild Editorial, the company editorial team responsible for these guides. AI assisted research and drafting. This is documentation-based guidance with proposed examples, not a claim of hands-on testing or an independent product review. Report a correction.
Cover: Rumeyda Tutak / Pexels, Pexels License. This is an illustrative source photograph, not an application screenshot or a generated test result.