Color grading is a post-production process common to filmmaking and video editing that alters the appearance of an image for presentation in different environments on different devices. It involves enhancing attributes such as contrast, color, saturation, detail, black level, and white balance, whether for motion pictures, videos, or still images. The terms color grading and color correction are often used synonymously, and the process can include generating artistic color effects through creative blending and compositing of different layer masks of the source image. In modern practice, color grading is performed digitally in a controlled environment, typically a dim or dark color suite, and is one of the most labor-intensive parts of video editing.
The origins of color grading lie in the photochemical film process known as color timing, which was performed at a film lab during printing by varying the intensity and color of light used to expose the rephotographed image. With this process alone, the user could not immediately view the outcome of changes, so a Hazeltine color analyzer was commonly used for real-time viewing of modifications. In the 2000s, with the increase of digital technology, color grading in Hollywood films became more common, shifting from photochemical to digital workflows.
Color Timing
Color timing is the earlier photochemical process used in reproducing film elements. Originally, "color grading" was a lab term for changing color appearance in film reproduction when going to the answer print or release print. By the late 2010s, this film grading technique had become known as color timing, involving changes to the duration of exposure through different filters during film development. Color timing is specified in printer points, which represent presets in a lab contact printer, where 7 to 12 printer points equal one stop of light. The number of points per stop varied based on negative or print stock and different presets at film labs.
In a film production, the creative team would meet with the lab timer, who watched a running film and made notes based on the team's directions. After the session, the timer would return to the lab and place the film negative on a Hazeltine device, which had preview filters with a controlled backlight, to pick exact settings of each printer point for each scene. These settings were punched onto a paper tape and fed to a high-speed printer, where the negative was exposed through a backlight to a print stock. Filter settings were changed on the fly to match the printer lights on the paper tape. For complex work such as visual effects shots, "wedges" running through combinations of filters were sometimes processed to aid the choice of the correct grading.
Telecine
With the advent of television, broadcasters quickly realized the limitations of live broadcasts and turned to broadcasting feature films from release prints directly via a telecine. This was before 1956, when Ampex introduced the first Quadruplex videotape recorder (VTR) VRX-1000. Live television shows could also be recorded to film and aired at different times in different time zones by filming a video monitor. The heart of this system was the kinescope, a device for recording a television broadcast to film.
The early telecine hardware was the "film chain" for broadcasting from film, utilizing a film projector connected to a video camera. As explained by Jay Holben in American Cinematographer Magazine, "The telecine didn't truly become a viable post-production tool until it was given the ability to perform colour correction on a video signal."
How Telecine Coloring Works
In a cathode-ray tube (CRT) system, an electron beam is projected at a phosphor-coated envelope, producing a small spot of light. This beam is scanned across a film frame from left to right, capturing the vertical frame information. Horizontal scanning is accomplished as the film moves past the CRT's beam. Once the photon beam passes through the film frame, it encounters a series of dichroic mirrors that separate the image into its primary red, green, and blue components. Each individual beam is then reflected onto a photomultiplier tube (PMT), where photons are converted into an electronic signal to be recorded to tape.
In a charge-coupled device (CCD) telecine, a white light is shone through the exposed film image onto a prism, which separates the image into the three primary colors. Each beam of colored light is projected at a different CCD, one for each color. The CCD converts light into an electronic signal, and the telecine electronics modulate these into a video signal that can then be color graded.
Early color correction on Rank Cintel MkIII CRT telecine systems was accomplished by varying the primary gain voltages on each of the three photomultiplier tubes to vary the output of red, green, and blue. Further advancements converted much of the color-processing equipment from analog to digital, and with the next-generation telecine, the Ursa, the coloring process became completely digital in the 4:2:2 color space. The Ursa Gold brought about color grading in the full 4:4:4 color space.
Color correction control systems started with the Rank Cintel TOPSY (Telecine Operations Programming SYstem) in 1978. In 1984, Da Vinci Systems introduced their first color corrector, a computer-controlled interface that manipulated color voltages on the Rank Cintel MkIII systems. Since then, technology has improved to give extraordinary power to the digital colorist. Today, many companies make color correction control interfaces, including Da Vinci Systems and Pandora International. Some telecines were still in operation as of 2018.
Color Correction and Enhancement
Some of the main artistic functions of color correction (digital color grading) include:
- Reproducing accurately what was shot
- Compensating for variations in the material (e.g., film errors, white balance, varying lighting conditions)
- Compensating for the intended viewing environment (dark, dim, bright surrounds)
- Optimizing base appearance for inclusion of special visual effects
- Establishing a desired artistic 'look'
- Enhancing and/or altering the mood of a scene, the visual equivalent to the musical accompaniment of a film (compare also film tinting)
Some of these functions must be prioritized over others; for example, color grading may be done to ensure that recorded colors match those of the original scene, whereas other times the goal may be to establish a very artificial stylized look. Traditionally, color grading was done towards practical goals, such as in the film Marianne, where grading was used so that night scenes could be filmed during the day.
Color Grading in AI Image and Video Generation
In the context of Generative AI and Artificial intelligence image and video generation, color grading is invoked as a visual style or photographic/cinematic technique within prompts. Prompt writers use specific keywords to guide models toward particular color palettes, contrast levels, and moods. Common terms include "teal and orange," "bleach bypass," "high contrast," "muted colors," "vintage film look," "cinematic color grading," and "color graded." These descriptors help neural networks produce outputs that emulate the look of professionally graded footage.
Models trained on large datasets of images and videos learn associations between textual descriptions and visual features, including color distributions. For instance, a prompt like "a futuristic city with neon color grading" might lead the model to generate images with saturated blues and pinks, while "a somber scene with desaturated color grading" could yield muted, low-contrast results. The effectiveness of such prompts depends on the model's understanding of color-related terms, which is shaped by training data and alignment techniques like RLHF.
Notable Practitioners and Tools
Historically, color grading was performed by skilled lab timers and later by digital colorists using specialized hardware and software. In the film industry, companies like Da Vinci Systems and Pandora International developed control interfaces that became industry standards. Modern software tools, such as DaVinci Resolve, Baselight, and others, are widely used in professional post-production. In the AI domain, researchers and engineers at organizations like OpenAI, Google DeepMind, and Anthropic have developed models that can apply color grading styles based on textual prompts, though specific practitioners are less commonly named.
Relationship to Other Concepts
Color grading is closely related to color histograms, which represent the distribution of colors in an image and are often used in AI models to analyze or adjust color. It also intersects with color normalization, a technique that standardizes color appearance across images, sometimes used as a preprocessing step in Machine learning pipelines. In Deep learning models, color grading can be seen as a form of style transfer, where the model learns to apply a particular visual style to generated content.
The process of color grading in AI generation is analogous to traditional post-production but is performed implicitly by the model rather than through explicit user control. Prompt writers act as virtual colorists, using language to steer the model's output toward desired aesthetics. This has led to a new vocabulary of color-related terms in prompt engineering, which is a key skill in prompt writing for generative models.
Challenges and Considerations
One challenge in AI color grading is achieving consistency across multiple frames in video generation. While still images can be graded individually, video requires temporal coherence to avoid flickering or abrupt changes. Models like those developed by Google DeepMind and other labs often use temporal smoothing techniques, but this remains an active area of research. Additionally, the interpretation of color grading terms can vary between models, leading to inconsistent results. Prompt writers often need to experiment with different phrasings and parameters to achieve the desired look.
Another consideration is the ethical and aesthetic implications of color grading in AI-generated media. As with traditional color grading, the choices made can significantly affect the mood and meaning of an image, and AI models may inadvertently perpetuate biases present in training data. Understanding the technical and artistic aspects of color grading is therefore important for both creators and consumers of AI-generated content.
Future Directions
As AI models continue to improve, the ability to control color grading through natural language is likely to become more precise. Advances in transformer architectures and attention mechanisms may enable models to better understand nuanced color descriptions. Integration with traditional color grading tools could also allow for hybrid workflows, where AI generates initial grades that human colorists refine. The ongoing development of large language models and their alignment with visual data will shape how color grading is invoked in AI prompts, making it an increasingly powerful tool for visual storytelling.