Complete chain
Not a collection of LUTs and nodes, but a sequence of known image states: capture/decode → input transform → scene-referred working space → grade → rendering/output transform → monitoring and delivery.
A good colour pipeline can be read like a route. Between any two stages you know the signal's gamut and tonal encoding, which operations belong there, and which transform is responsible for the next state.
Quick reference
RAW settings occur before the main node tree.
Technical normalisation of the source state.
Balance, match and look in a known working context.
Output transform + calibrated monitoring/delivery.
STATE MAP
A node by itself says nothing about colour state. The same CST can be an input transform, a local conversion or an output transform depending on placement. Document the states between tools, not just tool names: SOURCE AWG4/LogC4 → DWG/Intermediate → DISPLAY Rec.709/BT.1886. This notation quickly exposes missing and duplicate transforms.
RAW may exist before a chosen output gamut/gamma pair, while encoded ProRes log already arrives in a specific state. Camera RAW settings, debayering, white balance and decode gamma/gamut occur before the main node tree. If that stage changes, the downstream grade receives a different input even when the nodes are untouched.
The input transform should be technically repeatable. Its purpose is not to create a look but to interpret a camera/graphics encoding and move it into the chosen working representation. On a mixed-camera timeline, this is the first point where clips become comparable in mathematical context, although real differences in sensors, exposure and lenses remain.
Before display rendering, it is often useful to perform operations that should behave relative to scene exposure: exposure compensation, white-balance-like corrections, certain transforms and linear-light VFX operations. Scene-referred does not necessarily mean linearly encoded: log, Intermediate or ACEScct can be nonlinear working encodings while still representing scene relationships before display rendering.
A wide working gamut reduces the need to clip saturated colours early, and a suitable log-like transfer distributes precision and makes controls practical. But a huge gamut alone does not guarantee good results: operations must handle negative/out-of-gamut values correctly and effects must understand the timeline colour state. The working space is a production environment, not a publication format.
A film-emulation LUT, DCTL or look transform is only correct in the state for which it was designed. A LUT built for LogC3 should not be dropped blindly into DWG/Intermediate or after a Rec.709 output. A good practice is to separate technical normalisation, balance/matching and look into clear node groups and explicitly label the colour-state contract of each creative transform.
The output/rendering transform solves a problem that did not exist in the camera state: what contrast, peak luminance and gamut the target display can reproduce, and how scene values should be mapped into them. Tone scale, chroma/saturation behaviour, gamut mapping and display encoding form the output-referred image. This is why the same scene-referred grade can have different, valid SDR and HDR renderings.
A colourist judges not abstract RGB numbers but the result through a calibrated display path. OS colour management, clean video feed, DeckLink, monitor transfer/gamut, legal/full levels and HDR signalling can alter the image actually seen. The Resolve Viewer is useful, but it is not proof of the electrical/display signal leaving the system.
CG may be scene-linear ACEScg/EXR while titles and UI are display-referred sRGB/Rec.709. They should not be sent through the same camera input transform. VFX round-trips should define an interchange state and a viewing recipe: for example, scene-linear EXR remains unrendered while the monitor views it through the same Output Transform as the DI. Otherwise, a match inside one application does not guarantee a match on return.
If SDR, HDR and cinema versions split too early, creative changes must be synchronised manually. It is usually better to preserve a common scene-referred/master grade for as long as possible and create output-specific branches closer to rendering/delivery. Trim passes may still be required, but they become intentional deviations from a common master rather than three independent projects.
Exposure +1 stop before the Output Transform and lift/gain after it are mathematically different actions. Saturation before gamut mapping changes scene colourfulness; after mapping it changes rendered display RGB. Sharpen/blur can depend on gamma/linear state. Instead of saying 'this node must always be third', say 'this operation belongs in this specific state'.
Overly contrasty log after enabling RCM often indicates a double input/display transform. A dull Rec.709 clip in a wide timeline may have been tagged as camera log. Neon highlights shifting into odd hues can point to gamut handling rather than a 'bad camera'. Before creative correction, always verify the technical chain on a neutral/reference clip.
Record the project colour science, version, input mappings, RAW decode policy, timeline space, look placement, output transforms, monitoring calibration and delivery tags. For complex projects, add a diagram showing states and the owner of each transform. If another colourist can open the project a month later and reconstruct the intended signal path without guessing, the pipeline is well designed.
Label technical nodes by states rather than 'CST 1 / CST 2': INPUT AWG4/LogC4 → WORKING DWG/Intermediate and WORKING → OUTPUT Rec.709/Gamma 2.4. The pipeline can then be audited in minutes.
Technical definitions are anchored to standards bodies and primary specifications wherever that distinction matters.
General Input → scene-referred ACES → Look → Output architecture.
Rendering boundary from scene-referred to output-referred imagery.
RCM, Timeline Colour Space and practical CST workflow.
Specification of a typical Resolve working representation.
