AICineFocus

AI CineFocus was introduced with DaVinci Resolve 21 as an AI-assisted refocus and virtual-lens tool. Blackmagic describes it as a way to emulate realistic defocus and bokeh, including on phone footage where nearly everything was captured sharp.

The node builds a working depth model from a normal 2D image and uses that model to determine how far each region sits from the virtual focal plane. That is fundamentally different from a flat Gaussian Blur or a single roto mask.

The interface is organized around four mode tabs: Depth, Focus, Aperture and Result. Depth diagnoses and repairs the map; Focus selects the sharp region; Aperture visualizes the virtual-lens behavior; Result shows the final image while keeping all controls available.

In Depth mode, brighter areas are treated as closer, gray as mid-field and darker as farther away. If a face, hair, prop or background is assigned to the wrong depth, increasing aperture only makes the mistake more visible.

Depth Map Source = Internal generates the depth map inside CineFocus. Source = From Alpha Input uses an external map created elsewhere and fed through the alpha/depth input.

With external depth, verify polarity first. Use Invert when near/far are reversed. Blackmagic also specifically says Separate Layers should be 0.0 when an external Depth Map is used.

Adjust Map Levels exposes Far Limit, Near Limit and Gamma. These remap the depth range; they do not reconstruct geometry that the source depth map never contained.

Depth-map Post-Processing adds Post Filter, Contract / Expand, Blur and Separate Layers. These are repair tools for named boundary problems. Blackmagic warns that Depth Map Blur can itself create artifacts at image-object edges.

Focus mode marks in-focus regions in color and defocused regions in gray. The OpenFX split-prism overlay can be clicked to select a focus point or dragged to define a focus range.

Focus Distance is a normalized percentage based on the depth map rather than meters or feet. It is keyframeable, so a controlled rack/pull focus can be created after the shoot.

Depth of Field controls how broad the in-focus range is. It is distinct from Aperture Size: you can keep more of one subject sharp without globally weakening the near/far defocus.

Track Focus to Point ties focus to a viewer-overlay point that can be followed with the FX Tracker. Tracking does not fix a wrong depth estimate, so the full clip still needs review.

No Foreground Blur removes blur from foreground regions and may improve performance when the creative intent is to keep the nearest plane sharp. It should not be used as a blanket substitute for a correct map.

Bokeh Quality and Depth Map Quality are separate switches. The first controls defocus/bokeh rendering while the second controls internal depth estimation. Faster is useful for setup; Better/Enhanced are quality choices for later validation.

Aperture mode visualizes the internal lens model, including light-path spread before and after the focal point and bokeh slices at different depths. This makes it easier to understand what aperture controls are doing than tuning sliders blind.

Shaped Bokeh models blade count, rotation, curvature and pinwheel, while Anamorphic Lens stretches bokeh. These are lens-character decisions; arbitrary complexity does not make an image automatically more cinematic.

Lens Properties adds Chromatic Aberration and Spherical Aberration. They are best treated as subtle optical imperfections after the base depth/focus result already works.

Lens Distortion changes the virtual response toward frame edges. Curved Lens, Curvature, Lateral Aberration and Phase let bokeh and color spread become less mathematically perfect.

Foreground Edge Overlap addresses one of the hardest depth-aware blur problems: a near object becomes optically blurred and partly transparent, revealing background behind it. A 2D plate does not contain that behavior automatically, so CineFocus provides an overlap workaround.

Automatic Edge Handling creates defocused overlaps automatically and disables manual Overlap Boost. Compare the two methods on real difficult edges such as hair, fingers, foliage and thin props.

Remove and Replace Grain exists because real film grain or video noise is independent of scene depth. Simply blurring a grainy plate blurs the grain itself, which can make the post effect obvious.

CineFocus can remove grain before defocus, render a cleaner depth-aware blur, and restore texture afterward. Preview Grain and Threshold help separate noise from real detail, while Replace Grain After Defocus and Grain Replacement control restoration.

Additional Grain is a separate finishing stage with Strength, Size, Softness and Saturation. It is optional; adding synthetic grain when the source already has suitable texture can make the image worse.

Final QC must be temporal. Review focus pulls, hair/shoulder boundaries, overlapping objects, motion blur, bright bokeh, grain consistency and several representative frames. Many CineFocus failures only reveal themselves in motion or when subjects cross depth layers.