AI Glossary / Computer vision and multimodal AI / Beginner

Color grading

Definition

Color grading is the deliberate adjustment of an image's color and tonal appearance to support a consistent visual style, mood or storytelling purpose. In film, video and photography, it commonly includes balancing shots, shaping contrast and directing attention through selective color and brightness changes.

What it means

Two shots recorded moments apart can look different because lighting, exposure or camera settings changed. A grading workflow can make them feel visually connected and then develop an intended appearance, such as restrained neutrals or warm highlights. People often distinguish color correction, which addresses balance and consistency, from creative grading, which shapes the look. In practice, the work overlaps. Grading is an editorial choice applied to images, not proof that the resulting colors match the original scene. This distinction matters when images are used as evidence, product references or inputs to computer-vision measurements.

How it works

A controlled workflow first interprets the source image encoding correctly and establishes the intended output conditions. The colourist evaluates exposure, white balance and contrast, using calibrated monitoring and tools such as waveform displays or vectorscopes. Primary adjustments affect broad portions of the image, while secondary adjustments isolate particular colors or regions. Masks and tracking can keep a local adjustment attached to a moving subject.
Shots are compared within a sequence so that individual changes support continuity. Creative looks may use curves, channel operations or lookup tables, but their order and input assumptions matter. The result is reviewed through the intended output transform and delivery settings rather than judged only inside one editing interface.

Three practical examples

• A business video team matches interview shots recorded on different cameras so that cuts do not produce distracting changes in skin tone or contrast.
• A travel photographer creates a consistent series while retaining meaningful differences between sunrise, shade and indoor scenes.
• A film colourist uses a selective adjustment to reduce a distracting background color and guide attention toward the subject without changing the entire frame equally.

Think of it this way

Think of arranging the lighting and decoration of a room to create a particular atmosphere. The objects remain recognizable, but their presentation affects how the room feels. The analogy is limited because grading changes recorded pixel values after capture. It cannot physically relight every surface correctly or restore information that was never recorded.

Worked example

Suppose an illustrative normalized RGB sample is (0.40,0.35,0.30). A simple channel-gain operation multiplies red by 1.05, green by 1.00 and blue by 0.95, producing (0.42,0.35,0.285). This pushes that sample toward a warmer balance. A different sample may respond differently to a later curve or saturation operation, so one adjusted pixel does not define a complete grade.
The example is basic arithmetic, not a recommended preset or a perceptually uniform color operation. Actual results depend on whether values are linear, logarithmic or display-encoded, as well as the working space and output transform. Highlight values may exceed available output limits, and careless clipping can erase detail.

Illustrative case study

Fictional example: Harbor Learning edits an online course filmed over two days. One session has cooler lighting, so the editor first balances the shots using a reference and the intended viewing setup. A restrained creative grade is then applied across the sequence. During review, a colored diagram becomes harder to distinguish, so the team adjusts the treatment for teaching clarity. The case shows that a visual style should serve the material's purpose; it does not claim that one look suits every educational or business video.

Common misconceptions and mistakes

A lookup table is not a complete grading decision and may expect a specific input encoding. More saturation does not automatically make an image better. Matching numeric RGB values between differently encoded files does not guarantee matching appearance. Grading cannot reliably reconstruct clipped highlights or absent color information. A fashionable look may damage brand colors, diagrams or natural skin appearance if applied without review. Finally, a pleasing monitor preview is not sufficient evidence that all delivery versions will look consistent across their intended output conditions.

Business use

Businesses benefit from agreed reference looks, clear approval criteria and a repeatable color-management workflow. Decide whether the priority is faithful product representation, visual storytelling or both, and identify content that should remain constrained. Keep original media and editable grading decisions so later versions can be produced consistently. Review sequences rather than isolated attractive frames. For AI-generated and camera-captured assets used together, check that their tonal and color behavior remains coherent after final conversion, while avoiding edits that misrepresent a product's actual appearance.

People also ask

Is color grading only for movies?

No. It is used in photography, advertising, educational media and other image-based communication.

Are correction and grading completely separate stages?

They are useful conceptual distinctions, but practical workflows often combine balancing and creative decisions.

Can AI grade footage automatically?

Automated tools can suggest or apply adjustments, but source interpretation, continuity and the intended appearance still need evaluation.

Check your understanding

A preset looks attractive on one shot but makes a brand logo visibly incorrect in another. Should it be applied unchanged to the whole project?

Answer: No. The grade should be adapted and reviewed against the project's visual and accuracy requirements, including protected colors or content.

Related terms

• Color gamut
• Tone mapping
• colour correction
• white balance
• lookup table
• colour management
• image enhancement

Learn next

Learn exposure, white balance and basic color management before building elaborate looks. Then practice matching two shots, inspect scopes alongside the image and compare the exported result on the intended delivery setup.

More in Computer vision and multimodal AI

• Color difference (R−G, B−G) - Color difference signals R−G and B−G represent the red and blue channel values relative to the green channel at the same image location.
• Color filter array (CFA) - A color filter array, or CFA, is a spatial pattern of tiny optical filters associated with an image sensor's photosites.
• Color fringing - Color fringing is an unwanted colored border or halo around an image feature, often most visible at a high-contrast edge.
• Color histogram - A color histogram summarizes how frequently different colors, or ranges of color values, occur in an image or selected region.
• Color layout descriptor - A color layout descriptor is a compact representation of where broad colors occur within an image.
• Color models - Color models are mathematical ways of describing color through components or coordinates.

Sources

Blackmagic Design, The Colorist Guide to DaVinci Resolve 20: https://documents.blackmagicdesign.com/UserManuals/DaVinci-Resolve-20-Colorist-Guide.pdf Blackmagic Design, Color tools overview: https://www.blackmagicdesign.com/products/davinciresolve/color Official training and product documentation search excerpts consulted; full workflow and editorial review remain pending. Numerical adjustments are illustrative rather than product-specific instructions.

Previous: Color gamut Next: Color histogram