AIUltraSharpen

AI UltraSharpen arrived in DaVinci Resolve 21 as a new AI sharpening tool. Blackmagic describes it as an extremely high-quality sharpening process for moving video images rather than a conventional still-frame-only edge filter.

The key architectural idea is selective processing. The AI analyzes the picture and tries to sharpen only areas where added definition is useful instead of raising high-frequency contrast uniformly across the whole frame.

In Blackmagic's official example, a slightly soft face gains visible detail in the face and hair while the already blurred background remains largely untouched. That is a useful mental model: improve meaningful detail rather than making every pixel harder.

The node really does expose one dedicated control: Sharpen Amount. The captured Fusion Inspector shows 0.300000. That is evidence of one UI state and a convenient A/B reference, not a universal production preset.

Sharpen Amount sets how much sharpening you ask the tool to apply. The visible result still depends on what the AI interprets as useful structure, so the same numeric value can feel different on a face, fabric, foliage, graphics or low-resolution footage.

UltraSharpen may rescue a slightly soft or mildly missed-focus shot. Blackmagic explicitly says such material may be recoverable. That does not mean severe defocus can be reconstructed faithfully; once spatial information is genuinely absent, a larger Amount cannot recreate the original optical focus with certainty.

The tool can also intelligently improve the perceived resolution of low-resolution footage. In practical pipelines this makes it useful after an upscale or resize, when the target frame size is already established but local detail still reads soft.

UltraSharpen and Super Scale solve adjacent but different problems. Scaling changes spatial resolution and image dimensions; UltraSharpen is a finishing/detail-enhancement process. Do not treat one as a substitute for the other, and evaluate their order in the actual pipeline.

Check noise before sharpening. Any detail enhancer can make sensor noise, ringing, mosquito noise and compression blocks more visible. Selective AI behavior reduces blanket sharpening risk but does not remove the need for source-quality control.

After strong denoise, AI UltraSharpen can restore perceived crispness, but it can also emphasize plastic edges or residual cleanup artifacts. Balance cleanup and sharpening instead of compensating for one overly aggressive stage with another.

On skin, inspect pores, eyelashes, hair, lip edges and eye contours. A good result increases readability without crunchy microcontrast or synthetic-looking texture.

On text, UI and logos, inspect halos or ringing around high-contrast edges. Geometric graphics reveal excessive Sharpen Amount very quickly.

On foliage, fine fabric, fences and repeating patterns, inspect temporal stability. Detail enhancement that looks clean on one frame can crawl or shimmer in motion and after compression.

Effect Mask is a separate Fusion coverage input. It does not tell the AI which detail is important and does not change the analysis model; it only limits where the processed result contributes to Output.

A mask is especially useful when only a face, product, plate number, logo or screen needs extra definition while the background should not receive additional high-frequency energy. Tracking that mask remains a separate workflow when the subject moves.

If the source is already sufficiently sharp, UltraSharpen may add no useful information and can simply make the frame look digital. Use bypass as a real control condition rather than assuming an AI effect must improve every shot.

Judge sharpening at a useful Viewer scale, often around 100%, and then in the real delivery size. A reduced Viewer can hide halos, while extreme zoom can tempt you to fix pixel-level details the audience will never see.

Always play several seconds after UltraSharpen. Resolve 21 positions the effect for moving images, so a single still frame is not enough to certify temporal quality.

The captured 0.3 setting is useful as a repeatable A/B baseline: enable the effect, compare with bypass, then decide whether the shot needs less, more or zero. Do not copy 0.3 between shots as if it were a LUT-like preset.

A reliable production order is: identify softness/noise/upscale needs, add UltraSharpen, find the minimum sufficient Sharpen Amount, inspect edges/skin/textures at useful magnification, play the shot, add Effect Mask only when needed, then verify a representative export.