Dither

Purpose. Dither is an image filter in the Krokodove toolset integrated into Resolve 21. It is useful on smooth backgrounds, dark gradients, glows and motion graphics where limited color precision produces visible steps; it can also create intentionally crunchy retro texture.

The signal path is simple: MediaIn or a finished 2D comp → the yellow Dither Input → Output → MediaOut. This is a one-image processor, not a Merge or generator. The verified tool profile does not expose a separate mask, map or second image socket.

Diagnose the input first. If strong bands already exist upstream, Dither cannot recreate the missing tones. Compare the original in the Viewer, repair the cause or consider Deband before evaluating a new dithering pattern.

Output Bits Per Channel sets the per-channel quantization precision used by the effect. Reducing the number makes fewer levels available and generally makes quantization more obvious. This is not the export file bit-depth selector or a codec setting.

The captured Inspector shows Output Bits Per Channel = 8. Treat that as the captured interface reference, not a universal SDR delivery rule: the upstream working pipeline and the final output codec have their own precision boundaries.

Dither Method is one dropdown with four choices: None, Ordered, Simple Error Distribution and Floyd & Steinberg. These are four values of a single control, not four individual tools or four separate Inspector parameters.

None is the comparison mode without an added distribution pattern. Inspect its quantization at fixed Bits and Multiplier; it reveals why the other methods may be helpful on a gradient.

Ordered produces a regular spatial pattern. On skies or subtle fills the matrix can become conspicuous; for title art, retro graphics or a deliberately limited palette, that digital texture may be exactly the intended look.

Simple Error Distribution spreads error to nearby pixels rather than using the Ordered pattern. Compare it at 100% Viewer scale: the goal is to break up broad steps without drawing more attention to the texture than the gradient itself.

Floyd & Steinberg is a classic error-diffusion approach. Broad steps can become finer local variations in brightness or color; that is perceptual camouflage for quantization rather than restoration of missing sample values.

Multiplier adjusts the strength of the selected dither contribution. The capture shows 1.0, but the available Inspector evidence does not establish a universal internal scaling formula. Tune it after choosing Bits and Method, and inspect the result.

Color Space is on the Render tab. The captured menu lists RGB, HLS, YUV, YIQ, CMY, HSV, XYZ and LAB. These eight choices belong to one dropdown and may change how color or luma texture appears because the calculation uses a different component representation.

Render Color Space here should not be confused with a Color Space Transform node or Resolve project-wide color management. Picking LAB or YUV in Dither does not by itself deliver a properly color-converted master.

RGB is a sensible comparison baseline. If a saturated gradient develops objectionable chroma texture, compare YUV and LAB without also changing Bits, Method or Multiplier. No calculation space is automatically best for every input or codec.

For neutral technical dithering, graph order matters. Perform the main grade, Blur, Glow, Merge and final Resize upstream of Dither where possible. Downstream resampling or blur may destroy the fine spatial pattern and reveal steps again.

For art direction you may deliberately change that order: Ordered followed by Glow or Scale becomes part of a stylized image. Do not apply that creative sequence uncritically to clean delivery.

Dither and Film Grain are different tools. Grain is an aesthetic texture largely independent of target precision; Dither addresses discrete quantization levels. Decide whether the task is banding control or deliberate photographic texture.

A Fusion Viewer preview is not delivery approval. Inspect at 100%, play moving footage and render a short sample with the actual target codec. Compression may smooth fine noise away, create blocks or make temporal shimmer more visible.

Use representative test material: a neutral ramp, dark gradient, saturated fill and soft glow. Artifacts can look very different on each; an arbitrary camera still may hide them under natural sensor noise.

For graphics with transparency, also inspect the alpha and premultiplication edges. The captured controls establish the UI, not every internal rule governing auxiliary channels and alpha across Resolve builds.

A reproducible decision process starts with None at the intended precision, compares Ordered, Simple Error Distribution and Floyd & Steinberg at fixed Bits, then tunes Multiplier, tests Render Color Space and validates the actual encode.

Evidence boundary. Four original Inspector captures confirm all four controls and both dropdown inventories. Exact HDR, alpha, animated and codec-specific behavior needs a live test in the target Resolve version; passing a content audit is not a render test.