DaVinci Resolve remains the only NLE that ships with a dedicated, node-based color grading environment used by broadcast and theatrical colorists on actual finishing projects. Premiere Pro has Lumetri. Baselight has its own pipeline. Resolve has the Color Page — a tool that demands structured thinking about how corrections stack, compound, and interact. This guide covers the full grading pipeline from ingest to delivery with exact settings, honest limitations, and workflows tested on 40+ music video, commercial, and short film projects at BLKRIP Studio.
Resolve 19 ships with the Color Warper, HDR palette, AI-based Magic Mask, Magic Dust for noise reduction, and an overhauled node graph. The fundamentals — serial node flow, Lift/Gamma/Gain, Log wheels, Custom Curves — have not changed since Resolve 12. If you understand the node graph, you can grade in any version.
1. Node Pipeline Architecture
Serial, Parallel, Layer, and Outside Nodes
Every correction in Resolve lives inside a node. Nodes process sequentially from left to right by default (serial mode). The output of Node 1 feeds into Node 2, which feeds into Node 3. This serial chain is your primary organizational structure — and the one you should default to for 90% of corrections.
Parallel nodes split the signal into multiple branches that process simultaneously, then combine additively. Use them when you want to apply two independent corrections that should not interact — for example, one node for skin tone isolation and another for sky enhancement. The additive blending means both contribute to the final image without one "seeing" the other's adjustments. The drawback: parallel nodes are harder to troubleshoot because you cannot easily solo one branch to see its isolated effect.
Layer nodes stack like Photoshop layers. You assign a composite mode (Add, Subtract, Difference, Soft Light, Overlay) to each layer node and control opacity with the Key Output gain. Layer nodes are the correct choice when you need to blend a stylized look on top of a corrected base — the classic "grade on top of correction" split.
Outside nodes apply corrections to the inverse of the current qualifier or power window key. Add a qualifier on skin tones, then attach an Outside node — now your correction hits everything except skin. This is faster than duplicating nodes and inverting keys manually.
One honest limitation: Resolve's node graph does not support discontinuous routing. You cannot feed Node 5's output back into Node 2. Every graph is a directed acyclic tree. If you need iterative feedback loops, you are looking for Fusion, not the Color Page.
Standard 4-Node Music Video Grade
Node 01 — BALANCE: Lift/Gamma/Gain for exposure and white balance. Parade RGB scope. Target: waveforms centered, RGB channels aligned at the crossover point. Set Master Wheels: Lift -0.015, Gamma +0.008, Gain +0.025 as starting point for underexposed LOG footage.
Node 02 — CONTRAST: Custom Curves. Master curve: pull down shadows at 10% input to 5% output, lift highlights at 90% input to 95% output. Add a gentle S-curve with 3 control points. Do not crush blacks below IRE 5 or clip highlights above IRE 100.
Node 03 — LOOK: Hue vs Sat curve for teal-orange split. Pull saturation up 15% at the orange hue vector (roughly 30 degrees on the vectorscope). Pull saturation up 12% at the teal/cyan vector (roughly 195 degrees). Desaturate greens and magentas by 8-10% to tighten the palette.
Node 04 — FINISH: Soft power window vignette with feather 0.45. Add 0.8-1.2 density of Resolve's Film Grain (Kodak 5219 stock preset) if the footage is too clean. This node also holds any per-shot fixes: windowed sky replacements, local exposure bumps for faces.
Node Naming Convention (Do This From Day One)
- Label every node: BALANCE, CONTRAST, SKIN, SKY, LOOK, VIGNETTE, GRAIN. Double-click a node, press Ctrl+N (Cmd+N on Mac), type the name.
- Use node colors consistently. Set BALANCE nodes to blue, LOOK nodes to green, FIX nodes to red. Right-click a node > Node Color > pick.
- Group shots by scene. Select multiple clips in the timeline, right-click > Create Group. Apply shared corrections at the Group Pre-Clip level. Shot-specific fixes go at the Clip level. This two-tier structure saves hours on 200+ shot projects.
- Snapshots. Before any risky adjustment, right-click the clip > Grab Still. One click to revert. No undo history needed.
2. Primary Correction Workflow
Exposure Balancing with Parade RGB
Open the Parade RGB scope (Workspace > Scopes > Parade). The three waveform displays show Red, Green, and Blue channel levels from 0 (shadow) to 1023 (highlight in 10-bit). Your first job: get all three channels to start and end at the same points.
If the Red channel sits higher than Green and Blue, the image has a red cast. Lower Red Lift or raise Green/Blue Lift to align the shadow points. Then align the highlight points using Gain. The crossover — where all three channels meet in the midtones — tells you the white balance accuracy.
For LOG footage (S-Log3, V-Log, C-Log3), apply the official camera LUT first as a reference, then balance underneath it. Or skip the LUT entirely and grade natively using the Log Wheels, which work in LOG space and give you direct control over the LOG image without the clipping that LUTs introduce.
Quick test: toggle the LUT on and off (right-click clip > Bypass Color Grade). If the raw LOG image has visible color shifts between shots, fix those before applying any creative look. Painting over uncorrected LOG with a stylized grade produces shots that look good on your monitor but fall apart on phones and tablets.
White Balance Correction Using the RGB Mixer
The RGB Mixer gives you per-channel output control. For white balance correction, this tool is more precise than Temperature/Tint because it operates on the actual channel data rather than a perceptual model.
Method: find a neutral reference point in the frame — a white shirt, a grey card if you shot one, concrete, a white wall. Open the Waveform scope. Note which channel is high or low in the neutral area. Go to the RGB Mixer panel. If Red is high, lower the Red output to Green and Blue channels: set Red Output to 0.95 (or whatever the offset requires). Fine-tune by adjusting Blue Output up or down to match.
This approach is slower than dragging the Temperature slider, but it handles mixed lighting better. A scene lit by both daylight (5600K) from a window and tungsten (3200K) practicals will have different color casts in different areas. The Temperature slider applies a global shift. The RGB Mixer, combined with a power window isolating the window-lit area, handles mixed lighting correctly.
Limitation: the RGB Mixer is destructive in the sense that it reduces channel data. Aggressive RGB Mixer adjustments on 8-bit H.264 footage will produce banding. For 8-bit sources, use the Temperature/Tint controls or Color Warper instead — both maintain better mathematical precision on limited data.
Custom Curves vs Log Wheels for Contrast
| Parameter | Custom Curves | Log Wheels |
|---|
| Control granularity | Per-channel, unlimited points | Shadow/Midtone/Highlight per wheel |
| Ease of use | Steeper learning curve | Faster for basic adjustments |
| Precision | Highest — place control points at exact IRE values | Good for broad strokes |
| Shadow rolloff | Direct control via curve shape | Indirect — Low wheel affects 0-25% range |
| Highlight control | Direct — pinch or expand highlights independently | High wheel affects 75-100% range |
| Per-channel correction | Yes — separate R, G, B curves | Yes — separate wheels per channel |
| Best for | Film-like contrast with custom toe/shoulder behavior | Quick LOG-to-Rec.709 contrast mapping |
| Honest drawback | Easy to create posterization with too many points | Cannot create custom rolloff shapes |
Why the Color Warper Beats Lift/Gamma/Gain for Skin Tones
The Color Warper (introduced in Resolve 17) provides a mesh-based 2D hue/saturation grid. You click and drag specific hue regions without affecting adjacent hues. For skin tone adjustments this matters: the skin tone line on the vectorscope sits between orange and red-orange (the I-line). With Lift/Gamma/Gain, warming skin tones also warms everything in the same tonal range — brown furniture, wooden floors, warm clothing. The Color Warper isolates the specific hue vector.
Practical example: a subject's face reads too pink. In the Color Warper, grab the pink/red-orange region and drag it toward the orange/yellow-orange area. Adjacent reds and magentas stay put. With Lift/Gamma/Gain, you would need to add a qualifier on skin, which takes longer and often bleeds into similar-toned background elements.
The Color Warper's limitation: it operates in a fixed mesh resolution. Extreme pulls in one direction can create quantization artifacts in 8-bit footage. For RAW and 10-bit sources, this is a non-issue. For 8-bit H.264 from consumer cameras, keep Color Warper adjustments under 15% displacement.
3. Secondary Correction Techniques
HSL Qualifiers: Softness and Transition Settings
HSL qualifiers isolate a range of Hue, Saturation, and Luminance values. The H qualifier gets used 80% of the time. The S and L qualifiers handle edge cases — isolating bright saturated neon signs, or pulling a key on a dark blue suit that shares hue with shadows.
Settings that actually work:
Hue Softness: 20-35 for most natural keys (sky, grass, skin). Below 15 produces hard edges that no amount of softening can fix. Above 45 starts bleeding into adjacent hues and defeats the purpose of a secondary.
Saturation Softness: 25-40. Low-saturation keys (desaturated pastels, concrete, grey clothing) need higher saturation softness — 40-55 — because the qualifying range is narrow.
Luminance Softness: 15-30 for well-exposed footage. Underexposed footage with crushed shadows needs luminance softness of 35-50 to prevent the qualifier from flickering as shadow noise fluctuates frame to frame.
The Softness parameters control the transition zone between 0% and 100% key opacity. Think of them as feather on a mask. The actual Hue range is controlled by the center position and width of the H slider. Do not confuse the two — widening the H slider pulls in more hues, while increasing Hue Softness feathers the edge of whatever range you have selected.
Problems: HSL qualifiers struggle with compressed footage. H.264 4:2:0 chroma subsampling means color data is interpolated. A green screen that reads as solid green to your eye contains a 20-30 degree hue spread in Resolve's qualifier. BRAW, ProRes 4444, and R3D files key cleanly because the chroma data is not subsampled. If your qualifier edges look chewed, the source footage — not your settings — is the bottleneck.
Power Windows with Tracking
Resolve's Power Windows come in four shapes: Linear (gradient), Circle, Polygon, and Bezier curves. For face tracking, the Circle window works 70% of the time. Add a 0.40-0.55 softness value, position it over the subject's face, and enable the Tracker.
The Tracker offers three modes: Point, Cloud, and Frame. Point tracks a single reference point — fast but drifts when the tracked point rotates or scales. Cloud tracks multiple points — slower but handles rotation and scale changes. Frame uses full-frame analysis — slowest but most reliable for complex motion.
For talking-head footage with moderate head movement, use Point tracker with a reference point on the bridge of the nose. For walking shots with perspective changes, use Cloud with 5-8 tracking points distributed across the face. For shots with heavy camera movement plus subject movement, use Frame.
Track > stabilize > apply correction. If the track drifts, delete keyframes in the drift region, reposition the window, and re-track that segment. Manual keyframe correction is faster than trying to fix a bad track with more tracking.
The honest truth: Resolve's tracker is adequate but not best-in-class. SynthEyes and Mocha Pro produce more reliable tracks for complex shots. For 90% of color grading work (face windows, sky windows, simple object isolation), the built-in tracker is sufficient.
Magic Mask: Usage and Limitations
Magic Mask (Resolve 18+) uses neural network inference to isolate people and objects. It works. It also fails in specific, predictable ways.
When Magic Mask works well: single subject, clean background, consistent lighting, minimal motion blur. In these conditions, the mask tracks with 90-95% accuracy and saves 10-15 minutes per shot compared to manual qualification.
When Magic Mask fails: overlapping subjects, fast motion blur, low-contrast boundaries between subject and background, heavily backlit silhouettes, and any shot where the subject's edge detail is similar to the background (dark clothing against dark walls). In these cases, the mask flickers, drops frames, or bleeds into the background.
The processing cost is significant. Magic Mask renders at roughly 1-2 fps on an NVIDIA RTX 3080. For a 4-minute music video with 80 shots, generating Magic Masks for all shots takes 30-45 minutes of compute. Use Render Cache > User to pre-render Magic Mask nodes and restore real-time playback.
Workflow: apply Magic Mask on a dedicated node. Add a Blur (0.5-1.0 pixel) to the alpha output to soften the mask edge. Check every 10th frame manually. If the mask drops, switch to manual Power Windows for that shot. Do not ship a grade relying on unchecked Magic Masks.
4. Creative Grading Approaches
Film Emulation: FilmConvert Nitrate and Dehancer Pro
Resolve's built-in Film Look Creator (Resolve 19) provides basic film stock profiles. For production work, two third-party plugins dominate: FilmConvert Nitrate and Dehancer Pro.
FilmConvert Nitrate operates by matching your camera's sensor response curve to the target film stock's spectral sensitivity. It asks for your camera model (e.g., Sony A7S III, RED Komodo, Blackmagic Pocket 6K) and applies a scientifically modeled conversion. The grain engine generates per-frame grain that responds to exposure — brighter areas get finer grain, shadows get coarser grain, matching real film behavior.
Settings for a convincing Kodak Vision3 500T look: set FilmConvert to Kodak Vision3 500T stock. Set exposure to match your graded image. Grain amount: 35-50% for standard, 60-80% for heavy. Grain size: 1.0-1.5x. Enable Halation at 10-20% for the red halation bloom that occurs in real film highlights. Set Bloom to 5-10% for the highlight glow effect.
Dehancer Pro takes a different approach — it models the entire photochemical pipeline: negative exposure, development, printing, and projection. This gives you control over printer lights, fog, vignetting, and color cross-talk that FilmConvert does not offer. Dehancer's grain is perceptually more accurate but requires more tweaking.
Dehancer settings for Fujifilm Eterna 500: load the Eterna 500 LUT profile. Set Negative Density to 0.85-1.0. Print light adjustments: Red +2, Green 0, Blue -1 for a warm shift. Grain: Kodak 5219 profile, intensity 40%, size 1.2x. Enable Halation at 15%. Enable Gate Weave at 0.3-0.5% for subtle frame instability that sells the analog look.
Opinionated take: film emulation plugins are overused. Not every project needs grain and halation. A clean digital grade with intentional color design serves most commercial and corporate work better. Reserve film emulation for projects that explicitly reference analog aesthetics — music videos with retro themes, fashion films, narrative work set in pre-digital eras. The client who asks for "film look" usually wants contrast rolloff and warm shadows, not actual grain and halation. Hit those with Custom Curves and Lift warmth before reaching for a plugin.
Teal-Orange Complementary Split: Exact Parameters
The teal-orange look works because human skin falls naturally in the orange hue range (vectorscope I-line, 15-40 degrees). Placing the complementary color (teal/cyan, 180-210 degrees) in shadows and backgrounds creates a perceptual push-pull that draws the eye to faces.
Step-by-step in Resolve:
1. Balance the image first. The teal-orange look amplifies any white balance errors. Fix those in Node 01 before proceeding.
2. Node 02 — Hue vs Sat curve. Boost saturation at the orange hue vector by 12-18%. Boost saturation at the teal/cyan vector by 10-15%. Desaturate greens by 15-20% and yellows by 8-12% to remove hues that compete with the orange skin line.
3. Node 03 — Hue vs Hue curve. Shift shadow-range hues (below 40 IRE) 8-12 degrees toward teal. Shift highlight-range hues (above 70 IRE) 3-5 degrees toward warm orange. Use 3-4 control points for smooth transitions.
4. Node 04 — Lift/Gamma/Gain. Add +0.010 to +0.020 blue and -0.005 to -0.010 red in the Lift wheel (shadows go teal). Add +0.005 to +0.015 red and -0.005 to -0.010 blue in the Gain wheel (highlights go warm). Keep Gamma neutral — this is where skin lives, and shifting it creates noticeable color casts on faces.
5. Check skin tones on the vectorscope. They should stay on or near the I-line. If skin drifts toward magenta, back off the blue Lift. If skin goes too yellow, back off the red Gain.
The mistake most people make: pushing the split too hard. If the image looks obviously teal and orange, you have gone 50% too far. The best teal-orange grades are felt, not seen. Dial everything back by 30% from where you think it looks good.
Custom LUT Creation Workflow
01
Grade a reference shot to your target look using the full node pipeline. Choose a shot with skin tones, neutral elements, and a range of exposures — a medium close-up in good lighting works best.
02
Verify the grade on at least 3 different display devices: your grading monitor, a laptop screen, and a phone. LUTs amplify display-specific characteristics. If the grade only looks good on your calibrated monitor, the LUT will fail everywhere else.
03
Right-click the clip thumbnail in the Color Page > Generate LUT > 33 Point Cube. Save as .cube format. Resolve also exports .3dl and .mga but .cube is the most widely compatible format across NLEs and on-set monitors.
04
Test the LUT on 10 different shots from the same project. If 8+ shots look correct with the LUT applied and only need minor adjustments, the LUT is solid. If most shots need significant correction after the LUT, the LUT is too aggressive — rebuild it with milder settings.
05
Apply the LUT as the first node (or a Group Pre-Clip node) on all project clips. Grade on top of it in subsequent nodes. This gives every shot a consistent starting point while preserving flexibility for shot-specific adjustments.
06
Export a second version of the LUT at 65 Point Cube resolution for archival and high-precision applications. The 33-point version handles on-set monitoring and editorial preview. The 65-point version handles final color delivery where the extra precision prevents banding in smooth gradients.
5. HDR Grading
HDR10+ Metadata Handling in Resolve Studio
HDR10+ adds dynamic metadata to the HDR10 base layer — per-scene (or per-frame) tone mapping instructions that tell the display exactly how to render each moment. Resolve Studio has built-in HDR10+ metadata authoring since version 18.5.
Setup: In Project Settings > Master Settings, set Color Science to DaVinci YRGB Color Managed. Set Timeline Color Space to Rec.2100 ST.2084 (PQ). Set Output Color Space to the same. In the HDR palette (Color Page > HDR), set Max Fall (frame average light level) to 400-600 nits for most content. Set MaxCLL (maximum content light level) to 1000 nits for standard HDR delivery, or 4000 nits for premium theatrical content.
The HDR palette shows you the PQ curve from 0 to 10,000 nits. Most of your creative work happens in the 0-1000 nit range. The master wheel controls overall brightness. The Zone-based wheels let you target specific nit ranges: 0-100 nits for shadows, 100-400 nits for midtones, 400-1000 nits for highlights.
HDR10+ metadata authoring: After completing the HDR grade, go to Deliver page. Under Advanced Settings, enable HDR10+ Metadata. Resolve analyzes the graded timeline and generates per-scene metadata including MinPQ, MaxPQ, and MaxSCL values for each scene. Review the metadata in the HDR10+ panel — if any scene reports MaxCLL above your target ceiling, reduce specular highlights in that scene.
Key limitation: HDR10+ metadata does not change pixel data. It tells compatible displays how to adapt the PQ signal. Non-HDR10+ displays (standard HDR10 TVs) ignore the dynamic metadata and use static tone mapping, which may produce incorrect results. Always verify your HDR grade on both HDR10+ and standard HDR10 displays.
Information gain point: Resolve's HDR10+ metadata is authored at the deliver stage, not during grading. The colorist grades in PQ, then Resolve's metadata engine reads the graded pixel values and generates the appropriate tone mapping instructions. This separation means you can adjust the HDR10+ metadata without re-grading — useful when a platform rejects your deliverable for exceeding their nits ceiling.
Tone Mapping HDR to SDR
Every HDR project needs an SDR trim pass. Resolve provides two paths: automatic tone mapping via the Color Space Transform OFX, or manual SDR trimming.
Automatic: Add a Color Space Transform OFX on the last node. Set Input Color Space to Rec.2100 ST.2084, Output Color Space to Rec.709. Set Tone Mapping Mode to "DaVinci." This produces a usable SDR version in 30 seconds. It also crushes specular highlights and flattens the midtone contrast that you carefully built in the HDR grade.
Manual trim (the right way): Duplicate your HDR timeline. On the SDR copy, add a node after the HDR grade. This "trim node" adjusts specifically for SDR display limitations. Typical adjustments: reduce highlight rolloff by pulling the curve above 80 IRE down 5-10%. Add 3-5% contrast in the midtones to compensate for the reduced dynamic range. Increase saturation by 5-8% because SDR displays render color at lower perceived saturation than HDR displays at equivalent code values.
The manual approach takes 2-4 hours for a 4-minute music video. The automatic approach takes 30 seconds. For client work, the manual trim is worth the time — the difference is visible on any quality SDR display.
Dolby Vision Tools in Resolve Studio
Dolby Vision in Resolve requires a Dolby Vision license (separate purchase from Blackmagic, roughly $300 for the perpetual license). Once activated, Resolve adds a Dolby Vision palette alongside the HDR palette.
Dolby Vision uses Profile 5 (for streaming) or Profile 8.1 (for backward-compatible Blu-ray). Resolve handles both. The workflow: grade in PQ, then use the Dolby Vision trim controls to create the SDR backward-compatible base layer. The Dolby Vision metadata contains up to 254 analysis points per second, giving compatible displays extremely precise tone mapping.
Dolby Vision CM v4.0 (the current version) supports L2 (scene-based) and L5 (active area) metadata. Resolve generates L2 metadata automatically from the graded timeline. L5 metadata requires manual authoring for letterboxed content.
Practical reality: Dolby Vision is overkill for web content. YouTube does not support Dolby Vision. Netflix and Apple TV+ require it for original content. For most music video and commercial work, HDR10 is sufficient. Pay for the Dolby Vision license only when your deliverable spec explicitly requires it.
6. Noise Reduction
Temporal Noise Reduction: Frame Blend Settings
Temporal NR analyzes multiple frames to distinguish noise (random per frame) from detail (consistent across frames). The Motion Estimate Mode and number of reference frames determine the tradeoff between noise removal and motion artifacts.
Settings by footage type:
Talking head / interview (minimal motion): Motion Range: Small. Reference Frames: 5-7. Temporal Threshold: Luma 8-12, Chroma 12-18. This aggressive setting removes fine noise without visible artifacts because there is little motion to confuse the algorithm.
Music video / dance (moderate-fast motion): Motion Range: Medium. Reference Frames: 3-4. Temporal Threshold: Luma 5-8, Chroma 8-12. Lower thresholds prevent ghosting trails on moving limbs.
Action / sports (fast motion, camera movement): Motion Range: Large. Reference Frames: 2. Temporal Threshold: Luma 3-5, Chroma 5-8. Conservative settings are mandatory — temporal NR with aggressive settings on fast motion produces visible ghosting that looks worse than the original noise.
Always use the Better mode in Motion Estimation for critical work. Faster mode renders 2x quicker but produces visible edge artifacts in areas with fine detail movement (hair, fabric, foliage).
Spatial Noise Reduction: H.264 vs RAW
Spatial NR operates on a single frame. It blurs noise while attempting to preserve edges. The threshold and radius parameters control the balance.
For H.264 footage (4:2:0 chroma, typically 8-bit): Spatial NR is risky. The chroma blocks in H.264 footage respond poorly to spatial blur — the NR smooths the block boundaries, but the underlying data was never continuous. Results look like plastic. If you must use spatial NR on H.264, keep the threshold below 10 Luma / 15 Chroma, radius 3-5. Anything above produces wax-figure skin.
For RAW footage (BRAW, R3D, ARRIRAW): Spatial NR works well because the noise is Gaussian, not blocky. Threshold 12-20 Luma, 15-25 Chroma, radius 5-8 produces clean results that maintain texture. Apply before creative grading — noise interacts with color adjustments, and reducing noise first prevents the NR from fighting the grade.
For ProRes 422 HQ / DNxHR HQ: Treat these similarly to RAW for NR purposes. The data is robust enough that spatial NR with moderate settings (10-15 Luma threshold) cleans noise without destroying detail.
Resolution matters. 8K footage downscaled to 4K delivery hides noise through pixel averaging. A noisy 8K shot often looks clean at 4K output. Apply NR at the source resolution, then check at delivery resolution before committing.
When Noise Reduction Makes Footage Worse
Three scenarios where NR degrades the image:
1. Compressed noise. H.264 compression artifacts (mosquito noise around edges, macroblocking in shadows) are not Gaussian noise. Temporal and spatial NR treat them as noise and blur them, but the result is a smeared, soft version of the artifact. The footage looks simultaneously noisy and soft — worse than the original.
2. Film grain on scanned film. If you are grading actual film scans (16mm, Super 8, 35mm), the grain is the texture. Running NR on film scans removes the grain and leaves a flat, digital-looking image that defeats the purpose of shooting film. If grain reduction is needed for specific shots, use Dehancer Pro's grain management, which understands film grain structure.
3. NR before VFX compositing. If the footage goes to VFX after color, leave noise in place. VFX artists use noise patterns for tracking, edge detection, and compositing integration. Denoised footage produces worse VFX composites because the denoised edge data is less reliable for rotoscoping and keying. Apply NR as the last step before final delivery, not before handoffs.
7. Delivery and Export
Render Cache Workflow
Resolve's render cache pre-computes effects and stores the results as ProRes or DNxHR files on your cache drive. Three modes: None, User, Smart.
Smart mode lets Resolve decide what to cache based on playback performance. It works for simple timelines but makes poor decisions on complex node trees — it caches the wrong nodes and skips the heavy ones.
User mode is the professional choice. Turn on User mode (Playback > Render Cache > User). Then right-click individual clips or nodes and select "Render Cache > On." Resolve caches only what you specify. This gives you predictable cache behavior and avoids filling your cache drive with unnecessary files.
Cache location: set it to a fast SSD, not your system drive. Project Settings > General > Render Cache > Location. An NVMe SSD with 1000+ MB/s sustained write speed handles 4K cache files without bottlenecking. A spinning hard disk will bottleneck playback and defeat the purpose of caching.
Cache size management: a 4-minute 4K ProRes 4444 cached timeline consumes 40-60 GB. A 90-minute feature cached at 4K requires 1-2 TB. Clear cache files between projects. Playback > Delete Render Cache > All.
Before final render, always delete the render cache and render from source. The cache is a playback optimization, not a master. Cached files may contain errors from intermediate grading states that you corrected later.
Codec Selection for Different Deliverables
| Deliverable | Codec | Container | Bit Depth | Typical Bitrate | Use Case |
|---|
| Archival master | ProRes 4444 | MOV | 12-bit | ~500 Mbps at 4K | Maximum quality preservation, supports alpha channel |
| Broadcast delivery | ProRes 422 HQ | MOV | 10-bit | ~220 Mbps at 4K | Network TV, cable broadcast ingest standard |
| Web / social media | H.265 (HEVC) | MP4 | 10-bit | 30-80 Mbps at 4K | YouTube, Instagram, TikTok, Vimeo |
| Theatrical DCP | JPEG2000 | MXF (DCDM) | 12-bit | ~250 Mbps | Cinema projection, DCI-P3 color space |
| Streaming HDR | HEVC Main10 | MP4 / MKV | 10-bit | 20-50 Mbps | Netflix, Amazon, Apple TV+ HDR delivery |
| Editorial proxy | ProRes 422 LT | MOV | 10-bit | ~100 Mbps at 4K | Offline editing, fast scrubbing |
| VFX handoff | EXR (half-float) | EXR sequence | 16-bit float | ~200 MB/frame at 4K | Compositing in Nuke/After Effects |
DCP Export for Theatrical Delivery
Resolve Studio includes a built-in DCP (Digital Cinema Package) creator. Before Resolve 18, you needed a third-party tool like OpenDCP or EasyDCP.
DCP settings for standard theatrical: Resolution 4096x1716 (2.39:1 Scope) or 3996x2160 (1.85:1 Flat). Frame rate: 24fps (some European cinemas accept 25fps, but 24fps is the safe choice). Color space: DCI-P3 / XYZ. JPEG2000 compression at 250 Mbps.
In Resolve's Deliver page, select the DCP preset. Set the Reel Name (this becomes the film identifier in the cinema server). Set the Content Type: Feature, Trailer, or Advertisement. Set the studio name and rating. Resolve packages the DCP with the correct XML metadata structure.
Verify the DCP before shipping. Use a DCP validation tool — Quvis DCP Inspector (free) or Clipster (paid). Check the XML structure, the CPL (Composition Playlist), and the PKL (Packing List). A malformed XML in the CPL will cause the cinema server to reject the entire DCP, and you will not know until the projectionist tries to load it 30 minutes before the screening.
8. Performance Optimization
GPU Acceleration Settings
Resolve uses GPU for nearly all color page processing. CPU handles decoding, audio, and timeline management. The bottleneck is almost always GPU memory bandwidth, not GPU compute.
Configuration: Preferences > System > Memory and GPU. Set GPU Configuration Mode to Auto unless you have a specific reason to override. Resolve detects and uses all available GPUs — on a Mac with both M-series integrated GPU and an external eGPU, Resolve uses both simultaneously.
For multi-GPU systems: Resolve splits the frame into tiles and distributes them across GPUs. This helps with 8K footage but adds latency for 4K footage because the tiling overhead exceeds the parallel processing gain. If you are grading 4K, disabling all but the fastest GPU can improve real-time performance.
VRAM usage: a single 4K 12-bit frame consumes ~50 MB of VRAM. A 10-node grade on that frame needs 500 MB for intermediate buffers. At 8K, multiply by 4x — 2 GB per frame, 20 GB for a 10-node grade. An RTX 3080 with 10 GB VRAM will run out of memory on complex 8K grades. An RTX 4090 with 24 GB handles it.
On Mac: Apple Silicon unified memory architecture means VRAM equals system RAM. A Mac Studio M2 Ultra with 128 GB RAM has effectively 128 GB of VRAM. This is why Mac Studios dominate high-end Resolve suites — they handle 8K HDR grades that crash Windows workstations with consumer GPUs.
Proxy-to-Full-Reconform Workflow That Actually Works
Proxy workflows in Resolve break grades when the reconform is handled incorrectly. Here is the workflow that preserves grades across proxy-to-full-res switches:
1. Ingest with full-res media. Set Project Settings > Camera RAW to decode at Full Resolution. Add all source media to the Media Pool.
2. Generate proxies. Media > Generate Optimized Media. Set format to DNxHR LB or ProRes 422 Proxy at 1920x1080 (or 1280x720 for 8K source). Resolve writes proxy files to the Cache location.
3. Enable proxy playback. Playback > Proxy Mode > Half Resolution or Quarter Resolution. Resolve swaps source files for proxy files transparently — the timeline, edits, and grades remain linked to the original full-res clips.
4. Grade in proxy mode. Every node, qualifier, window, and keyframe is stored as metadata pointing to the original clip. The proxy mode only affects what you see during playback — the underlying metadata references full-res media.
5. Switch to full-res for final render. Playback > Proxy Mode > Off. Resolve reads the original source files, applies the grade metadata, and renders at full resolution.
This works because Resolve's grade metadata is resolution-independent. Qualifier keys are stored as HSL ranges, not pixel coordinates. Power Windows are stored as parametric shapes, not bitmaps. When you switch from proxy to full-res, Resolve recalculates every operation at the new resolution.
What breaks: Magic Mask. Magic Mask generates pixel-based alpha channels at the proxy resolution. When you switch to full-res, the alpha channel does not recalculate — it gets upscaled, producing soft edges. Solution: apply Magic Mask only after switching to full-res, or re-render Magic Mask nodes at full resolution before final export.
Another breakage point: Bitmap windows. If you create a bitmap-based power window by drawing on the frame at proxy resolution, it upscales poorly. Use parametric windows (Circle, Polygon, Bezier) whenever possible — these are resolution-independent.
Proxy Workflow for 8K Source Media
8K footage (RED V-RAPTOR 8K, Sony VENICE 2 8.6K, Canon EOS R5 8K RAW) demands a structured proxy workflow. Decoding 8K RAW at full resolution requires 4-6 GB/s of memory bandwidth per stream. Two streams for a dissolve exceeds most workstation capabilities.
Practical proxy settings for 8K:
Generate Optimized Media at 1920x1080, DNxHR LB, 24fps matching the timeline frame rate. This produces files roughly 1/16th the size of the 8K source. A 128 GB CFexpress card of 8K RED footage generates 8 GB of proxies.
If your workstation can handle it, generate at 3840x2160 DNxHR SQ for a better proxy that allows semi-accurate evaluation of detail and focus. This costs 4x the storage but gives you useful feedback on sharpness and fine detail during the grade.
Set the Cache location to a separate physical drive from the source media and the project database. Three separate drives: source media on RAID, cache/proxy on SSD, project database on NVMe. This prevents I/O contention during playback.
Never skip the proxy step with 8K. The temptation to "just play back at half resolution" from the RAW source exists — and on a fast enough workstation, it works for simple timelines. But add 8+ nodes of grading, noise reduction, and film emulation, and the decode pipeline becomes the bottleneck. Proxies decouple decode performance from grading performance.
Render Cache Strategies by Project Type
Music video (3-5 min, 50-100 shots, 6-12 nodes per shot): Use Smart cache during the creative exploration phase. Switch to User cache for final playback verification. Cache only shots with NR, Magic Mask, or heavy Color Warper usage. Estimated cache size: 20-40 GB at 4K.
Commercial (30-60 sec, 5-15 shots, 10-20 nodes per shot): Use User cache from the start. Cache every shot — the short timeline means cache generation takes under 5 minutes. Delete and regenerate cache after each major grade revision. Estimated cache size: 5-10 GB at 4K.
Feature film (90-120 min, 1500-3000 shots, 4-8 nodes per shot): Use User cache selectively. Cache only problem shots that drop frames during playback. Generate full cache overnight before client review sessions. Estimated cache size: 500 GB - 1 TB at 4K.
Documentary (variable length, 500-2000 shots, 2-6 nodes per shot): Smart cache handles most documentary timelines because the node trees are simpler. Cache shots with heavy NR (interview footage shot at high ISO). Estimated cache size: 100-300 GB at 4K.
9. Built-In Film Grain vs Composite Application Grain
When to Add Grain in Resolve vs After Effects or Fusion
Resolve's Film Grain effect (OpenFX > Film Grain) generates procedural grain with stock presets: Kodak 5219, Kodak 5248, Fujifilm 8553, and generic coarse/fine options. It is fast, renders in real time on most GPUs, and produces grain that responds to the image's luminance — shadows get coarse grain, highlights get fine grain.
Add grain in Resolve when: the grain is the final creative touch on a finished grade, the grain needs to be consistent across the entire timeline, and the project does not involve VFX compositing. This covers 80% of use cases — music videos, commercials, and most narrative work.
Add grain in Fusion or After Effects when: the grain needs to interact with VFX elements. For example, if you are compositing CG elements onto live-action footage, the grain must be applied after the composite so that CG and live-action elements share the same grain pattern. Adding grain in Resolve before the Fusion composite produces a mismatch — the live-action plate has grain but the CG element does not (or has different grain).
Add grain in the composite application when: the grain needs to be resolution-dependent in a specific way. Resolve's Film Grain generates grain at the timeline resolution. If you are delivering at multiple resolutions (4K theatrical master and 1080p web version), the grain will look different at each resolution. Adding grain at the composite stage lets you generate resolution-specific grain per deliverable.
Practical recommendation: for a standard music video or commercial grade with no VFX, use Resolve's Film Grain on the last node. Set the grain stock to Kodak 5219, density 0.8-1.2, and size 1.0x. This is the most common grain setting for professional music video delivery. For anything involving VFX, leave grain for the composite stage.
10. Honest Limitations and Practical Advice
Things Resolve's Marketing Will Not Tell You
- Resolve's free version does not support noise reduction (temporal or spatial), HDR grading, film grain, or GPU-accelerated scopes. The free version is a learning tool, not a production tool. Resolve Studio ($295 perpetual) is the minimum for professional work.
- The Color Page is not designed for assembly editing. Cutting in the Edit Page and grading in the Color Page is the intended workflow. Attempting to grade in a disconnected timeline without proper EDL/XML conform will produce misaligned grades when the edit changes.
- Resolve's audio tools are adequate for monitoring but not for final mix. Route audio to Pro Tools or Logic Pro for any project where audio quality matters. The built-in Fairlight mixer has improved significantly but still lacks the plugin ecosystem that Pro Tools offers.
- Color management in Resolve 19 defaults to DaVinci YRGB Color Managed, which is not ACES. ACES is available as an option but requires manual configuration and breaks compatibility with some OpenFX plugins. If your pipeline involves VFX handoffs, test the ACES workflow before committing the project to it.
- Resolve's LUT export is limited to 33-point and 65-point 3D LUTs. If you need higher precision (128-point LUTs for scientific applications), you need a third-party tool like Lattice or Colorfront.
- The scopes in Resolve are mathematically accurate but the UI rendering introduces slight inaccuracies. For broadcast-legal QC, use an external waveform monitor (Tektronix, Leader, or Flanders Scientific) rather than relying on Resolve's built-in scopes.
Quick Reference: Keyboard Shortcuts for the Color Page
Essential Shortcuts (Mac / Windows)
| Action | Mac | Windows |
|---|
| Add serial node | Opt+N | Alt+N |
| Add parallel node | Opt+P | Alt+P |
| Add layer node | Opt+L | Alt+L |
| Add outside node | Opt+O | Alt+O |
| Delete node | Opt+Backspace | Alt+Backspace |
| Disable/enable node | Ctrl+D | Ctrl+D |
| Grab still | Ctrl+Alt+G | Ctrl+Alt+G |
| Toggle before/after | Ctrl+D (on corrected clip) | Ctrl+D (on corrected clip) |
| Reset all nodes on clip | Ctrl+Alt+R | Ctrl+Alt+R |
| Copy grade from previous clip | Alt+Ctrl+C then Alt+Ctrl+V | Alt+Ctrl+C then Alt+Ctrl+V |
| Zoom scope to region | Alt+drag on scope | Alt+drag on scope |
| Toggle Proxy Mode | Playback > Proxy Mode toggle | Playback > Proxy Mode toggle |