Dehaze

Resolve 21 lists Dehaze under Resolve FX Color and marks it Studio Version Only. On the DaVinci Resolve Fusion page it can live inside a normal 2D node graph; Resolve FX are not available as native nodes in standalone Fusion Studio.

Dehaze targets visible aerial perspective such as smog, airlight and atmospheric haze that reduce contrast and color separation with distance. The filter automatically generates a simulated depth map from the frame, applying more corrective color/contrast adjustment to regions it estimates as farther away and less to regions it estimates as closer.

That depth matte is inferred, not measured camera Z. Dehaze receives no scene geometry, depth pass or Camera3D, so sky, water, backlight, reflections and low-contrast surfaces can be classified incorrectly. Display Depth exists to inspect this hypothesis before trusting the final image.

Dehaze Strength is a coupled correction, not a plain Contrast slider. Raising it subtly increases contrast, especially in shadows, rebalances color toward the complement of the selected Haze Color and selectively intensifies saturation. Lowering it reduces contrast, shifts color toward Haze Color and selectively reduces saturation, producing a more haze-like result.

Haze Color identifies the smog/airlight/haze color to minimize. A poor sample changes the direction of the entire color compensation, so sample an actually veiled distant region rather than a local colored object, specular highlight or saturated surface.

Display Depth reveals the internally generated simulated depth matte. Resolve 21 explicitly recommends turning it on before adjusting Shadow and Highlight so the near/far estimate can be judged directly.

Shadow and Highlight are not ordinary tonal shadow/highlight controls. Shadow raises or lowers the darkest parts of the simulated depth map, which represent the areas estimated to be farthest away. Highlight raises or lowers the lightest parts, which represent the areas estimated to be closest.

The Fusion Effect Mask is a separate spatial gate. It can restrict where Dehaze contributes, but it does not replace the internal simulated depth matte; near/far weighting is still estimated by Dehaze inside the admitted region.

Every Resolve FX also exposes a host Blend control. Treat it as the final mix of an already-correct result, not as a way to hide a bad Haze Color or broken depth matte: establish Haze Color, inspect/tune the matte, set Strength, then trim Blend or compare bypass.

Use Dehaze only when the problem is genuinely atmospheric. A flat shot caused by log gamma, exposure, lifted blacks, flare or a low-contrast grade should have that technical problem corrected first; forcing Dehaze onto the wrong problem can create dirty shadows, color casts and unnatural plane separation.

Treat Dehaze as a model of atmospheric veil rather than a global contrast enhancer. If low contrast comes from another cause, the algorithm can reinforce the wrong regions.

The internal simulated depth matte is derived only from the current 2D frame. It does not know scene geometry, lens distance, object distance or a real Z-buffer.

Display Depth is the primary diagnostic view for the model. Inspect it before pushing Strength because Strength scales a correction that already depends on the near/far hypothesis.

Light and dark values in Display Depth should not be reduced to a normal luma-mask interpretation. They are part of Dehaze's internal depth hypothesis and must be read as near/far weighting.

Shadow and Highlight shape the extremes of the simulated depth matte and should not be used as ordinary tonal shadow/highlight controls. Tune them from the diagnostic view.

If Dehaze misclassifies sky as near or a dark foreground object as far, correct the depth weighting before increasing Strength.

Haze Color models the color of airlight. It affects the direction of color compensation rather than merely tinting the final frame, so a bad sample can push neutrals across the image.

A useful Haze Color sample usually comes from a relatively uniform distant veil where the visible color belongs to the atmosphere rather than a subject, highlight or local light source.

Red/Green/Blue are components of one Haze Color and are not independent source-channel grading sliders.

Dehaze Strength couples several visual changes: contrast, color separation and saturation response. One adjustment can therefore look like several corrections at once.

Positive Strength generally reduces haze-like appearance while lowering it can move the image toward a hazier state. This remains the same atmospheric model rather than a separate fog generator.

Excessive Strength can produce crushed shadows, oversaturation and unnatural foreground/background separation, especially when simulated depth is already wrong.

Effect Mask and simulated depth operate at different levels: the mask says where the effect is allowed, while internal depth says how the correction is weighted inside that region.

A gradient Effect Mask is not a depth map. It only varies contribution of the final effect while near/far weighting remains Dehaze's internal estimate.

Display Depth travels through the normal image Output as a diagnostic representation; it does not create a separate mask/Z socket or change physical topology.

Dehaze can be fragile on sky, water, reflections, flare, backlight and low-texture regions because monocular depth inference is ambiguous there.

On faces and skin, aggressive Dehaze can create unpleasant local contrast and saturation even when the distant plane improves. A localized mask can be more appropriate.

On drone and landscape footage, judge Dehaze during camera motion because a wrong depth estimate can flicker between adjacent frames even when one still looks plausible.

If a shot has already received a strong creative grade, Dehaze may interpret intentionally altered contrast/color relationships as depth cues. Placement relative to grading should therefore be deliberate.

With log/flat material, decide which signal state Dehaze should analyze. There is no guarantee that placement before and after CST/OCIO produces the same depth hypothesis.

Running Dehaze before basic exposure/white-balance normalization can feed it technically distorted RGB, while running it too late can feed it stylized contrast.

The shared Resolve FX Blend should be a final mix, not a way to hide a bad Haze Color or broken depth matte. If the effect only works at very low Blend, recheck the model first.

Strong Dehaze does not restore information that was physically lost in haze. It redistributes contrast/color based on an estimate rather than reconstructing hidden scene detail.

Dehaze is not a replacement for selective color correction. If only hue/saturation needs adjustment, ColorCorrector/Curves may be more causally transparent.

Before Merge, Dehaze analyzes one plate/branch; after Merge, it analyzes the assembled composite. Added foreground or UI can materially change the internal depth estimate.

Transparent graphics, titles and CG foreground can confuse near/far estimation after Merge. Atmospheric treatment of a plate is often safer before those elements are added.

Dehaze is a Studio-only Resolve FX; standalone Fusion Studio is not a universal execution environment for it. That is a deployment constraint rather than a footnote.

A live target build is still required to validate host-specific UI, Common Settings, serialized IDs and performance. Static documentation defines semantics but does not replace runtime certification.

For automation, record the exact Resolve build and save a minimal .setting/.comp with Dehaze because visible labels do not guarantee stable internal IDs.

Performance and temporal stability should be checked at target resolution and real shot duration, not from a single Viewer frame.