Representation
Where scene-light relationships end and the rendered image begins: scene-linear vs scene-referred, output/display-referred state, the rendering transform and practical operation ordering in Resolve.
Scene-referred numbers describe relationships in the original scene before display rendering. Display-referred numbers describe an image already shaped for the capabilities and conditions of a particular display. The Output Transform is the boundary between those meanings.
Quick reference
Values describe scene exposure/colour relationships.
This is an encoding property, not a synonym for scene-referred.
Data is oriented to target display/viewing conditions.
This is where scene values become a display image.
REPRESENTATION
A scene-referred representation preserves a relationship to relative scene exposure/colour before the system decides how the scene should look on a particular display. A bright window may be many times above 18% grey without yet being required to become 100 or 1000 nits. Camera-linear, LogC4, ACEScct and DWG/Intermediate are different encodings, but they can all exist within the scene-referred part of a pipeline.
Linear describes the mathematical relationship between a code value and a measured light quantity: doubling the light doubles the linear value. Scene-referred describes what the data means relative to the scene. ACEScg can therefore be scene-linear, while ACEScct or DaVinci Intermediate can be scene-referred but nonlinearly/log-like encoded. This distinction is critical for compositing and operation choice.
After colour rendering, scene values are mapped into the range and gamut of a target display. Black, diffuse white, highlights and saturation now have meaning within the display system and viewing conditions. A Rec.709/BT.1886 master or PQ HDR deliverable is typically an output/display-referred representation. When you modify it, you are working on the rendered image rather than directly on the source scene-light relationships.
The move from scene to display can include a tone scale, adaptation/viewing assumptions, saturation/chroma changes and gamut mapping. ACES explicitly defines colour rendering as mapping scene colour-space coordinates into output-referred image data while accounting for viewing-condition differences and reproduction limits. An Output Transform can therefore change appearance even with perfect primary matrices — that is its job.
Before the Output Transform, +1 stop can scale a scene-exposure relationship and allow the rendering transform to remap highlights. After the output transform, a similar gain scales display RGB and may simply raise black/white, clip the signal or break the intended contrast. The two may look similar in midtones but diverge strongly in highlights and saturation.
Camera white balance and chromatic adaptation operate on interpretation of the source/scene state. If a strong WB error is corrected only after Rec.709 rendering, channels may already be clipped or nonlinearly compressed and the original relationship may no longer be recoverable. Display-referred RGB can still be creatively tinted, but that is not the same technical task.
Before output mapping, saturation may create scene colours far outside the target gamut, relying on later gamut compression. After output mapping, saturation moves already rendered display RGB and reaches the gamut boundary/clipping more quickly. The same 'Saturation = 60' in two states is therefore not the same operation.
In a manual CST pipeline it is useful to separate PRE-OUTPUT and POST-OUTPUT visually. Balance, exposure, camera matching and a scene-referred look usually occur before the Output CST/DRT; display trims, legalisation, output-specific graphics or small finishing corrections may intentionally live after it. The rule is not absolute — what matters is knowing which side of the rendering boundary each operation occupies and why.
UI, logos and web stills are often authored directly for sRGB/Rec.709 display. Forcing them through a camera-style Input Transform plus scene rendering can alter contrast, white and saturated brand colours. In a managed project, use correct input classification or a separate branch for display-referred assets. Scene-replacement/VFX plates are different: their pipeline may be scene-linear and must match the camera plate.
Rec.709 SDR and PQ/HLG HDR have different luminance capabilities and viewing expectations. Simply re-encoding one scene-referred master from Gamma 2.4 to PQ without a new rendering decision does not create a proper HDR image. A multi-delivery pipeline must reconsider tone scale, diffuse/reference white, highlight allocation and gamut mapping for each target.
Some workflows attempt to bring display-referred media back into a scene-like working context through an inverse DRT/CST. This is useful for mixed assets, but if the original display master has already been clipped, quantised or gamut-limited, inverse mathematics cannot recover the original camera latitude. The result is a convenient working representation reconstructed from the display image, not the original scene.
You cannot physically 'see scene-referred log as it is': a monitor always emits a limited range of light. When a Viewer shows log as flat, that is merely one display mapping of its code values without the intended rendering. Correct evaluation of a scene-referred grade requires a viewing/output transform and a calibrated display target.
Draw an imaginary vertical line through the Output Transform on your node graph. To the left are scene-referred decisions; to the right are display-specific decisions. That alone removes much of the confusion around CST and look-node ordering.
Technical definitions are anchored to standards bodies and primary specifications wherever that distinction matters.
Definitions of colour rendering, Input/Output/Look Transforms and scene-referred concepts.
Scene-referred common encoding and rendering toward displays.
Transition from scene-referred ACES to an output-referred display image.
Practical scene-referred RCM and transform placement in Resolve.
