calesthio/generative-media-skills/color-grading-finishing

Color grading and finishing direction

Use this skill to guide the color and finishing stage of a video project, whether the material is fully AI-generated, camera-originated, motion graphics, screen capture, avatar footage, product renders, or a mixed-source edit.

Treat color as both a technical pipeline and a storytelling surface. A good finish makes shots belong together, preserves required colors, supports the intended emotion, survives compression and platform conversion, and gives downstream editors/compositors unambiguous instructions.

Start by separating the job

Before proposing fixes, classify the work into three layers:

• Color correction: normalize the image so shots are usable and consistent. Balance exposure, white balance, black/white points, contrast, saturation, and obvious color casts. Documented fact: Blackmagic describes balancing color as the first correction step, using primary tools to adjust brightness and levels and remove unwanted tints before creative grading begins (Blackmagic Design Resolve Color page, verified 2026-07-10).
• Color grading: create the intended look after the material is technically balanced. Shape mood, palette, contrast style, warmth/coolness, density, texture, highlight behavior, and scene-to-scene emotional progression.
• Finishing: prepare the final master for review and delivery. Check legal/video levels, gamut, color tags, SDR/HDR target, accessibility contrast, caption/graphic readability, compression resilience, safe areas, artifact repair, and handoff materials.

Do not call a creative look "correction" when it intentionally changes color. Do not hide a technical mismatch behind a look; a stylized grade still needs matched shots and a valid delivery transform.

State evidence quality in your output

When giving color direction, separate:

• Documented facts: standards, platform specs, delivery targets, official tool behavior, or measured QC results. Cite or name the source when consequential.
• Empirical observations: what you saw in the current footage, frames, scopes, test encodes, or reference stills.
• Production heuristics: practical choices that often work, but are not universal rules.

Example phrasing: "Documented delivery fact: YouTube recommends BT.709 for SDR uploads. Observation: the generated close-up shifts from magenta to yellow between frames 81-105. Heuristic: match the product shot to the packshot first, then bend the environment toward the campaign look."

Decide the color-management path before grading

Color decisions are only meaningful inside a known transform chain.

Ask or infer:

• What are the sources? Camera log, raw, display-referred Rec.709, generated MP4, EXR/PNG sequence, screen capture, animation, product render, or unknown social download.
• What is the target master? SDR Rec.709, HDR PQ, HDR HLG, P3 theatrical/display, web sRGB, or multiple deliverables.
• Is the grade happening in a managed workflow? ACES, DaVinci Wide Gamut/Intermediate, scene-linear compositing, display-referred timeline, or unmanaged NLE filters.
• Which transform is technical and which is creative? Input transforms, display/output transforms, and conversion LUTs are not the same thing as a look LUT.
• What will be reviewed on? A calibrated reference display, client laptop, phone, social preview, or mixed review environment.

Documented facts to preserve:

• ITU-R BT.709 defines HDTV production/exchange parameters, including D65 white, Rec.709 primaries, 16:9 1920x1080 image format, and 8/10-bit nominal code ranges (ITU-R BT.709-6 PDF, verified 2026-07-10).
• ITU-R BT.2100 specifies HDR-TV production/exchange parameters using PQ or HLG methods (ITU-R BT.2100-3 PDF, verified 2026-07-10).
• ACES Output Transforms convert scene-referred ACES image data to a rendered state for a specific output device and viewing condition; do not treat the display transform as optional decoration (ACES documentation, verified 2026-07-10).
• OpenColorIO is a production color-management system used in motion picture, VFX, and animation pipelines, and is commonly used to keep applications aligned (OpenColorIO, verified 2026-07-10).

Production heuristic: if the material is generated MP4 with no reliable metadata, treat it as display-referred until proven otherwise. Avoid "log conversion" LUTs on AI-generated clips unless the source was explicitly generated or exported in that log/color space.

Make a grading brief before touching controls

A useful grading brief is short and operational:

• Target deliverables: e.g. 9:16 SDR Rec.709 social master, 16:9 YouTube SDR, HDR hero cut plus SDR trim pass.
• Reference state: links or stills for palette, contrast, skin tone, product color, black level, highlight feel, and acceptable grain/noise.
• Must protect: skin, brand colors, product materials, packaging, UI colors, legal disclaimers, medical/food/beauty claims, accessibility contrast.
• Scene families: interview, product macro, generated b-roll, screen capture, motion graphics, avatar, archive/UGC.
• Look intent: 3-6 adjectives tied to controllable attributes. "Premium, clean, warm" is vague; "warm key light, neutral whites, soft shoulder highlights, restrained saturation outside product red" is actionable.
• Review scope: correction-only pass, look-development stills, full sequence match, platform proof, final QC, or revision notes.

If the user asks for "cinematic," translate it into attributes: deeper contrast? softer highlights? warmer shadows? lower saturation? film-like halation? green/cyan bias? textured grain? Do not assume one universal cinematic look.

Correction workflow

Use this order unless a project-specific pipeline requires otherwise:

1. Conform and tag. Confirm frame rate, resolution, color space tags, data/video range, source order, and whether stills/graphics have embedded profiles.
2. Normalize sources. Apply correct input transforms or technical conversions. Remove accidental double transforms. Put sources into the working color space before judging the look.
3. Primary balance. Adjust exposure, black level, white level, contrast, white balance, and global saturation. Use scopes for objective checks; displays and ambient viewing can mislead.
4. Shot match. Pick a hero shot per scene, match adjacent cuts to it, then check the whole sequence in thumbnail/lightbox or side-by-side review. Match luminance and contrast before hue nuance.
5. Secondary fixes. Protect skin, product, sky, food, UI, clothing, or brand elements with hue/sat/luma qualifiers, masks/windows, curves, or object tracking when needed. Documented fact: Resolve-style secondary grading selects parts of an image by hue, saturation, or luminance; windows can isolate objects spatially (Blackmagic Design, verified 2026-07-10).
6. Look pass. Apply the creative palette and contrast behavior across matched shots. Use stills to compare variants.
7. Artifact pass. Fix banding, chroma noise, flicker, generated texture shimmer, edge halos, over-sharpening, compression blocks, and color discontinuities.
8. Delivery/QC pass. Check scopes, gamut, levels, color tags, caption/graphic contrast, platform crop/safe zones, export settings, and a test upload/transcode when relevant.

Shot matching for generated and mixed-source edits

For AI-generated clips, expect sources to vary even when prompts are similar. Match by perceptual continuity, not by pretending every shot came from the same camera.

Prioritize:

• Cut continuity: the viewer should not feel an unintended exposure, white balance, or saturation jump at the cut.
• Subject priority: faces and products should match across shots before backgrounds do.
• Scene logic: a sunset shot may be warmer than an office shot; a mismatch is only a problem if it breaks the scene's intended continuity.
• Graphic integration: lower-third colors, product labels, and UI overlays should remain stable across graded footage.
• Motion consistency: flicker across frames is a different problem from shot-to-shot mismatch; inspect playback, not only still frames.

Documented research context: video generation literature treats spatial consistency, including color and shape stability across frames, and temporal consistency as central generation problems; inconsistency can produce flicker and distortions (arXiv survey on spatiotemporal consistency, verified 2026-07-10). Production implication: if color changes every few frames, a static grade may not solve it; request regeneration, frame-level repair, deflicker, or replacement.

Skin, product, brand, and claim-sensitive color

Protect colors that carry identity, trust, safety, or legal meaning:

• Skin: keep natural hue relationships unless the style intentionally departs. Watch for green/magenta contamination, over-smoothing, excessive orange saturation, crushed facial shadows, and highlight clipping on foreheads/cheeks.
• Product: match approved packshots, swatches, material references, or client-provided stills. Do not make a product look like a different colorway, finish, or ingredient.
• Food and beauty: avoid grades that make food look spoiled, cosmetics look like a different shade, or skin claims appear artificially exaggerated.
• Medical, safety, and finance: preserve warning colors, UI status colors, disclosure text, and any color-coded meaning.
• Brand palettes: protect logos and brand surfaces with isolated corrections or overlays. A global teal/orange look that shifts a brand blue into cyan may be unacceptable even if the shot looks attractive.

Production heuristic: when product color accuracy matters, build the look around the product rather than applying a look globally and trying to rescue the product afterward.

Look development

Create looks with decisions, not just adjectives:

• Contrast architecture: high key vs low key, deep blacks vs lifted blacks, hard shoulder vs soft rolloff, bright whites vs restrained highlights.
• Palette: warm/cool bias, complementary palette, monochrome accents, neutral product field, selective saturation, seasonal color cues.
• Density: airy and open, dense and premium, documentary neutral, glossy commercial, horror/sci-fi contaminated, pastel social.
• Texture: clean digital, film grain, halation, bloom, diffusion, gate weave, VHS/archive, crisp product render.
• Attention control: guide the eye with luminance, saturation, and contrast rather than only vignettes.

Use references as constraints:

• A look reference communicates palette and mood.
• A technical reference communicates black/white levels, highlight rolloff, skin/product target, or HDR/SDR behavior.
• A do-not-copy reference may be useful when the user likes one element but not the whole finish.

Do not claim a look will exactly match a feature film, camera stock, or brand campaign unless you have the original source, display target, and a reasonable matching process.

LUTs and transforms

Be precise about LUT roles:

• Technical transform LUT: converts from one known color encoding to another. It requires the correct source and destination assumptions.
• Creative look LUT: imposes a style. It may clip, compress, or shift colors; use it as a starting point, not an unquestioned truth.
• Show LUT / viewing LUT: aligns review and production around a repeatable intended display rendering.

LUT cautions: