DeepToImage

Resolve 21 marks the Deep Image Compositing Toolset as Studio Version Only. A Deep Image can hold multiple samples at different depths inside one screen pixel, which lets dMerge, dHoldout and other Deep operators solve occlusion before the result is reduced to a flat frame.

DeepToImage performs that reduction: it flattens the depth samples and outputs a conventional 2D image. In production it normally sits after every operation that still needs Deep structure. Standard image tools can follow it, but downstream Deep operators cannot reconstruct the original sample stack from the flattened result.

Blackmagic explicitly notes that Deep compositing requires a linear colorspace. DeepToImage does not automatically repair an incorrect color pipeline. In particular, Source Color Space on the Image tab does not perform a Gamut conversion; it records or supplies source-space metadata that downstream tools can use.

The Controls tab has only two Deep-specific controls. Flip Depth inverts the depth information. Treat it as a correction or diagnostic for a mismatched Z convention, not as an artistic layer-order control. Volumetric Composition controls how overlapping Deep samples are flattened: enabled, they are blended as semi-transparent volumes, which matters for fog, VDBs and other volume renders.

Volumetric Composition is not a generic 'higher quality' switch. Opaque surface samples may show little difference, whereas genuinely overlapping semi-transparent samples can composite incorrectly when the option is disabled. Decide from the data you are flattening, not from a universal preset.

Image > Process Mode is Fusion field processing. Full Frames is the normal progressive choice; NTSC/PAL-HD field modes and reversed variants exist for actual interlaced workflows. This control does not change Deep flatten quality or sample ordering.

Image > Depth is easy to misread because of the name. Here Depth means the pixel color depth of the flattened 2D output—Default, int8, int16, float16 or float32—not the Z depth stored in the Deep Image. Keeping float16/float32 is usually safer when HDR or substantial VFX grading continues downstream.

Source Color Space is metadata, not a Color Space Transform. Resolve 21 says it can add source-space metadata when none exists so a downstream Gamut node or Saver can use it. If a real colorspace conversion is required, do not substitute this menu for an explicit color-management stage.

Source Gamma Space and Remove Curve define a separate transfer-function boundary. Remove Curve removes the selected gamma curve or performs the corresponding log-to-linear conversion, effectively producing a linearized result. Because Deep compositing is intended to be linear, this can be important—but enabling it on data that is already linear is equally wrong.

Pre-Divide / Post-Multiply belongs to alpha representation handling around processing. Treat it as a straight/premult technical boundary, not as a generic halo fix. Diagnose the actual RGB/alpha representation of the source before changing it.

The physical contract is intentionally simple: one required orange Deep Image input and one ordinary 2D Image output. There is no need to invent an Effect Mask, Camera or second image input. The complexity lives in the Deep data being flattened, not in a large port count.

**A Z pass is not a Deep stream.** An EXR carrying flat RGBA plus one depth channel does not automatically hold multiple samples at each pixel. Check the actual data domain before connecting: DeepToImage cannot fabricate Deep samples from a flat Z pass.

**dMerge before, ColorCorrector after.** dMerge combines actual Deep streams and retains depth-aware sample data; an ordinary ColorCorrector after DeepToImage works on the flattened RGBA result. If a Deep holdout is needed later, flattening is placed too early.

**Linear processing matters more than Viewer appearance.** Deep blending and transparency need a coherent linear-light workflow. Gamma/display-encoded RGB can look familiar in a Viewer while producing mathematically incorrect blends between depth samples.

**Metadata is not color conversion.** Image > Source Color Space provides source-space information for downstream tools but does not replace Gamut or a Color Space Transform. Changing that selection alone must not be described as altering the actual RGB pixel values.

**Curve Type works together with Remove Curve.** Auto uses available transfer metadata, Space selects a curve explicitly, and Log exposes log-related controls. Only when the source truly needs linearization should you choose the right curve and enable Remove Curve.

**RGB appearance cannot establish Z direction.** Test Flip Depth on a Deep shot with known surface ordering, comparing the same frame with the option on and off. If the incoming Z convention is correct, flipping it may introduce incorrect occlusion.

**Volumetric mode needs volume content.** Two opaque surfaces may show little difference when Volumetric Composition is toggled. Its main purpose is overlapping semitransparent volume samples such as fog or VDB renders; inspect overlap accuracy rather than general contrast.

**Pixel bit depth and scene depth are different.** Image > Depth selects the channel precision of the resulting 2D frame. Float16/float32 can preserve HDR values and gradients after flattening; they cannot increase the number of original Z samples or fix occlusion.

**Field processing is not Deep processing.** Process Mode deals with progressive/interlaced field handling, not Deep quality. Reversed field modes change field order, not Z direction; only the separate Flip Depth control addresses that depth interpretation.

**Inspect Alpha after flattening.** At boundaries between opaque and translucent samples, examine RGB and Alpha separately. Pre-Divide / Post-Multiply helps with straight/premultiplied representations but cannot reconstruct lost Deep geometry or repair a fundamentally wrong source.

**Inspector captures document controls, not recommended defaults.** The saved Controls screenshot shows both checkboxes selected. This records one UI state, not a rule that every shot must use Flip Depth and Volumetric Composition. The exercises define their A/B states explicitly.

**Verification boundary.** The illustrated Flow diagrams validate Deep → 2D connection types, but static checking cannot verify real EXR samples, volumetric blending, transfer curves or Resolve renders. Those require a separate native Studio test with controlled frames.