CopyAux

**CopyAux bridges RGBA and auxiliary data.** A Fusion image can carry standard Red/Green/Blue/Alpha plus Z, Normal, Vector, World Position and IDs. CopyAux moves whole channel groups; it does not identify objects semantically or manufacture a missing 3D pass.

**Two Mode directions.** `Aux to Color` exposes a chosen auxiliary group through RGBA for viewing or 2D processing. `Color to Aux` stores RGBA back in a selected auxiliary group after optional math. These modes do not represent two independent image inputs.

**Data is not display.** A negative Z coordinate or negative component of a normal is valid mathematical data, even though a conventional Viewer interprets RGB in display range 0–1. A black viewer after `Aux to Color` alone does not prove the auxiliary pass is empty.

**Channel-group packing.** One component maps to `aaa1`, two to `ab01`, three to `abc1`, four to `abcd`. Z therefore appears as `zzz1`, UV as `uv01`, and Normal as `nx ny nz 1`. Alpha follows the group-packing convention; it is not necessarily copied from the beauty image.

**Preserving Z depth.** Fusion often uses negative eye-space Z, becoming more negative farther from the camera. Copying such values into integer RGBA can collapse nearly the entire range to zero. Keep depth in float32 unless an explicit viewing or delivery remap is intended.

**Output precision.** `Match Aux Channel Depth` adjusts RGBA depth to the auxiliary source, `Match Source Color Depth` preserves the existing color depth, and Force float/int overrides it. For Z, Normal, Vector and Position, integer output without remapping risks clipping and quantization.

**Object and material identifiers.** Object ID identifies an assigned object and Material ID identifies an assigned material. Neither is already a binary matte or an object color palette. Isolating a particular ID requires an explicit value selection after exposing the pass.

**Texture, normals and motion.** Texture carries UV, Normal carries surface direction XYZ, Vector/Back Vector carry temporal offsets, World Position stores scene coordinates, and Disparity stores stereo offsets. They are not interchangeable or inherently display-ready RGB passes.

**Missing Channel: Fail.** The `Fail` policy reports absent passes instead of silently producing substitute pixels. For EXR, inspect Loader mapping; for Renderer3D, inspect enabled output channels. Prefer failure to silent fallback when depth, IDs or motion vectors are essential.

**Missing Channel: Use Default Value.** Missing groups default to zero except Z, whose documented fallback is `−1e30`. This is a technical sentinel, not a measured object depth or a meaningful scene estimate. Treat it explicitly in diagnostics and subsequent math.

**Kill Aux Channels does not compress color.** After copying into RGBA, it discards remaining auxiliary groups, potentially reducing cache footprint. Before enabling it, check whether later tools still need Z, Vector, IDs or Coverage. This choice can permanently remove downstream access to those passes.

**Remapping is conditional, not a Mode.** The original Inspector shows `Enable Remapping` and a dynamic label such as `Color to Color`. Enabling it exposes From/To ranges and related actions. It remaps numeric values before copying; it is not a third direction of the Mode switch.

**Fixed ranges beat Viewer auto-scaling.** `From Min/Max` specify the input domain and `To Min/Max` the output. For example, Normal [−1,1] can be displayed over [0,1], or a selected Z range [−1000,0] can be mapped to [0,1]. These ranges are deliberate choices; out-of-range values need clipping review.

**Detect Range versus Update Range.** Detect Range scans the current image and replaces From boundaries with detected extrema; Update Range expands an existing interval to include newly discovered extrema. Running Detect Range independently every frame may make the display jump; lock a stable range for motion.

**Invert and reversed From bounds.** Conditional Remapping can invert a value mapping, and reversing From Min/Max can also change its direction. This is data-domain inversion, not a generic artistic Negative: it changes the interpretation of measured depth or vectors and needs explicit review.

**Round-trip is not guaranteed lossless.** `Aux to Color → processing → Color to Aux` is a legitimate data path, but destructive grading, premultiplication, clamping or reduced bit depth make restoration lossy. Compare the initial and restored auxiliary group numerically, not merely by appearance.

**Mask and image are distinct inputs.** The manual documents a required `Input` (2D image) and optional `Effect Mask` that limits CopyAux after processing. A Rectangle mask does not carry an auxiliary Z pass and cannot replace the required AUX-bearing image stream.

**Renderer3D and EXR are sources, not guarantees.** An EXR file or Renderer3D connection does not ensure every auxiliary pass is populated. Check renderer output selections, Loader mapping and actual channel inventory. CopyAux can only copy what enters its Input.

**Alpha and cache QC.** Packing one, two or three components may produce an output alpha of 1 by channel-group convention, rather than original transparency. On transparent beauty passes, separate auxiliary visualization, final RGBA compositing and the requirements for retaining auxiliary data.

**Acceptance boundary.** Four preserved Inspector images evidence Controls and menu choices; Blackmagic documentation explains Mode, depth, Missing and Remapping. The port model and teaching graphs can be validated statically, but actual Resolve 21 / Fusion Studio output, temporal behavior and bit accuracy still require native tests.