ChangeDepth

ChangeDepth [CD] receives one 2D image and passes it downstream in the selected channel representation. Controls contains exactly two tool-specific menus: Depth and Dither. Five depth entries and three dither modes are confirmed by the three preserved Inspector captures.

Depth = Keep avoids forcing a conversion at this point in the graph. It does not mean the upstream stream is necessarily float32 or that earlier nodes worked at maximum precision.

int8 and int16 are integer channel representations; float16 and float32 are floating-point formats. Channel depth and delivery encoding are different layers of the pipeline. Choosing float16 cannot create details absent from an 8-bit input.

Converting HDR or a smooth gradient to integer precision too early can introduce quantization banding or lose out-of-range values before subsequent Glow or ColorCorrector operations. Keep floating-point precision through demanding compositing stages when practical.

Dither primarily matters when reducing precision. None disables masking; Error Diffusion spreads rounding errors to neighboring samples; Additive Noise introduces low-level noise that makes regular steps less apparent. Neither restores information removed by quantization.

Assess dithering at 100% Viewer zoom on smooth tonal ramps and semi-transparent edges, across several frames. The monitor, playback and exported codec may have separate artifacts, so isolate the stage at which the bands appear.

Orange Input accepts a 2D image, optional blue Effect Mask gates where the processed result contributes, and a single Output remains an image stream. Masking does not produce independently stored bit depths in different image regions.

ChangeDepth does not convert gamma/gamut, resize a frame, choose an export codec, or control an MP4 bitrate. Debug the full signal chain rather than automatically inserting this node when a final encoded file has bands.

Fusion Studio manuals describe depth conversion as one way to reduce intermediate memory use. Resolve documentation separately notes 32-bit-float rendering: do not promise int8 as a universal Resolve performance optimization without measuring the target build.

The two Controls dropdowns are documented against the three original screenshots. The Settings page, physical port layout in a live Flow, and pixel-level output on Resolve 21.1 have not been directly verified in this editorial pass.