Automatic Film Dirt & Dust Removal: The Complete PFClean Restoration Guide
- 3 hours ago
- 29 min read
This guide provides a professional overview of automatic film dirt & dust removal in PFClean, covering the complete range of restoration approaches from fast, automated cleaning through to advanced precision control using multi-parameter settings. Whether you are managing a large-scale archive digitisation programme, preparing high-volume film transfers, or restoring a single valuable and historically significant reel, this guide explains how to select the right workflow for your material and preservation objectives.
This guide is part of a wider collection of PFClean tutorials and restoration resources covering film restoration, digital preservation, and workflow planning.
Key takeaways:
Remove dirt after stabilisation and de-flicker, before final colour and grain work, detection accuracy depends on a stable, balanced frame.
Use Dustbust for fast, uniform dust at volume; Auto Dirt Fix when you need size ranges, channel control, or spatial/temporal switching.
An infrared defect map, where the scanner provides one, is the gold standard, it replaces detection with measured emulsion data.
Heterogeneous damage wants a multi-pass stack: one targeted effect per defect type, fine to coarse.
Every repair needs grain matching and a blend radius, a technically correct patch without them reads as a dead zone.
Table of contents
1. Introduction
Dirt, dust, airborne debris, and emulsion chips are among the most common defects in archival film material. Every stage of a film's life, production, processing, projection, storage, and scanning, introduces the possibility of foreign matter settling onto the emulsion or the gate, or physical damage to the emulsion surface itself. The result in a scan is a frame-by-frame scatter of bright or dark anomalies that can range from a faint sparkle of fine dust to large opaque chunks of print debris.
Emulsion chips are a distinct category of damage. Unlike airborne dirt, which sits on the film surface, emulsion chips occur when fragments of the emulsion layers are physically broken away, by rough handling, brittle stock, or abrasion in a gate or projector. Because colour film builds its image from stacked dye layers, a chip that removes one or two layers leaves behind a sharp-edged fragment with a pronounced colour bias, most commonly a vivid blue or green, rather than the neutral dark or bright appearance of ordinary debris. This colour signature is a useful detection handle, but it also means standard composite detection can struggle to isolate chips cleanly, a challenge addressed in Section 12.
Beyond the visual disruption to the viewer, untreated dirt actively interferes with other restoration processes. Grain analysis, colour balance sampling, and temporal matching are all confused by random high-contrast anomalies. Cleaning the dirt first gives every downstream algorithm, and every human operator, a faithful, unobstructed view of the actual image.
PFClean offers three main automatic tools for dirt and dust removal, each optimised for a different working scenario:
A streamlined, high-speed effect for rapid removal of fine dust and sparkle with minimal setup.
A full-featured effect offering complete control over detection and repair, capable of addressing everything from fine grain-level dust to large print debris and coloured emulsion chips.
Handles dirt and dust as part of a dedicated optical-scanning simulation workflow and is covered briefly in its own section below.
Choose your method based on the nature and severity of the damage and how much time you have to calibrate the result.
2. Effect Ordering
When using the Workbench effects Dustbust or Auto Dirt Fix, place them after any stabilisation, auto de-flicker, and colour balancing effects in the processing stack. The reason for this ordering is that Workbench effects operate on the results of the effects that come before them. By ensuring the frame is already stable, free from flicker, and correctly balanced, the dirt and dust detection process becomes more accurate, allowing the fixes to integrate more naturally and cleanly into the image.
As a general workflow rule, remove dirt and dust after stabilisation but before final colour work and grain management. Position your chosen dirt removal effect below any Stabilize and De-Flicker effects in the Workbench stack. If using the Dustbust to remove small debris, use this before the Auto Dirt Fix so the detection algorithm in the Auto Dirt Fix can focus on the larger debris.
Why this matters:
Workbench effects process the results of the effects that come before them, so the quality of the input image directly affects the accuracy and cleanliness of the repair. Placing Dustbust or Auto Dirt Fix after stabilisation, de-flicker, and colour balancing ensures the detection algorithms are working with a consistent and predictable image.
Stable, non-flickering frames improve detection accuracy. Dirt detection relies on comparing the current frame with surrounding frames and identifying anomalies. If the image is shifting due to gate weave, jitter, or flicker, stationary details can be mistaken for moving dirt, while genuine defects may be harder to isolate. A stabilised and de-flickered image provides a consistent spatial reference, allowing the detection process to work more effectively. We have a complete guide to stabilisation here to help you get started.
Balanced frames allow cleaner defect removal. Colour inconsistencies and exposure fluctuations can affect how defects are identified, particularly when dirt or dust appears as high-contrast anomalies. Applying colour balancing before dirt removal helps create a more consistent image for detection, allowing the corrections to blend more naturally with the surrounding image information.
Note on ordering within a multi-pass stack: If you are stacking multiple Auto Dirt Fix instances see Section 9, order them from the most targeted correction (such as fine dust and small defects) to broader corrections (such as larger debris). Each effect processes the output of the effects above it, so carefully ordering passes helps prevent later, broader passes from re-detecting or altering already corrected areas.
3. Digital Wet Gate — Automatic Cleaning
Type: Inline Toolset | Location: Digital Wet Gate | Complexity: Easy
When to use this
The Digital Wet Gate (DWG) is PFClean's dedicated pipeline for simulating an optical wet-gate scanning process. It is the appropriate choice when you are processing material that was originally intended for wet-gate scanning, particularly footage with surface scratches and physically-embedded dirt, and you want cleaning to be handled as part of a unified optical simulation rather than as a separate Workbench pass.
Within the DWG's Defect Manager, the Dirt/Dust controls provide automatic detection of bright or dark debris marks that appear for a single frame at a time.
How it works
The DWG Dirt/Dust detector analyses frame-to-frame differences to identify transient pixel changes that do not belong to genuine image motion. Detection works on either Bright, Dark, or Both types of marks simultaneously.
Key controls
Detect % — The minimum change in pixel intensity from frame to frame required for a pixel to be classified as a defect. Lowering this value detects fainter marks; the default is 10%. Reducing it too far will increase false positives in areas of high motion.
Reject % — A confidence filter that discards low-certainty detections before applying fixes. The default is 50%. Increasing this value removes more uncertain areas, trading completeness for precision.
Show Defects overlay — Enables a red pixel overlay in the Cinema window showing all currently detected dirt/dust pixels. Use this actively when calibrating Detect % and Reject % values.
Show Difference overlay — Displays how much each pixel has changed relative to the original. Neutral grey indicates no change; brighter or darker pixels indicate significant alteration. Use this after processing to verify that repairs are integrating naturally.
IR Defect Map — If the material was scanned with an infrared defect channel, the DWG can use this data directly instead of frame-difference detection.
When to use another effect
The DWG Dirt/Dust controls are intentionally streamlined: they are designed to operate as part of the broader DWG pipeline. For material requiring granular per clip control over detection size ranges, channel isolation, spatial versus temporal fix switching, or grain-matched patching, use Auto Dirt Fix in the Workbench instead.
4. Dustbust — Fast Spot Removal
Type: Automatic | Effect: Dustbust | Complexity: Easy
When to use this
Start here for high-volume material with fine, relatively uniform dust and sparkle. Dustbust is optimised for speed: it performs motion-compensated detection and fix at a significantly higher throughput than Auto Dirt Fix, trading granular control for rapid turnaround.
It is the natural first pass when you are processing archive transfers at scale, or when a quick review confirms that the damage is limited to small, bright or dark spots with no complex coloured artefacts, large debris, or emulsion damage requiring precise thresholding. You can also use the Dustbust effect as the first of multiple passes on a complex clip, for example it can be stacked above an Auto Dirt Fix to remove the fine sparkle and debris before passing it to the Auto Dirt Fix which can be calibrated to remove more complex larger issues.
Dustbust is not suited to large chunks of print debris, emulsion chips, or material where detection needs to be restricted by size range or channel.
Key controls
Type — Select Dark, Light, or both simultaneously. Match this to the polarity of the damage visible in your material. Most dust on film appears as bright (light) spots when transmitted through a scanner; negative dust and debris often appears as dark.
Detect / Fix channels — Specify which RGB channels are used to detect spots, and which are fixed. In many scans, the blue channel carries more noise than red and green. Restricting detection to red and green channels only, while still fixing all three, significantly reduces false positives in noisy material without sacrificing repair coverage.
Lookahead — The number of frames before and after the current frame examined to confirm a spot is transient. A spot must be present in the current frame and absent in the before/after frames to be classified as a defect. The default value of 1 is appropriate for standard single-frame dirt. Increasing this allows detection of persistent marks that last two or more frames.
Thresh % — The minimum intensity difference required to detect a spot. Increasing this value restricts detection to only the most prominent (brightest or darkest) marks. Decreasing it captures fainter dust at the risk of increased false positives in areas of complex motion.
Dilate — Expands the detected area by the specified number of pixels before the fix is applied. A small dilation value (1–2 pixels) is often beneficial to ensure the repair fully covers the edges of each spot. Be cautious: excessive dilation combined with a low Reject % can cause a detected false-positive area to suppress legitimate image detail.
Reject % — Controls how aggressively the effect discards detections it considers uncertain. Increasing this value reduces the risk of removing image detail that has been misidentified as dirt, at the cost of leaving some genuine marks untreated.
Motion — Sets the proxy resolution used for motion compensation during detection. Higher resolution settings improve detection accuracy in areas of complex or fast motion, at the cost of processing time. For clean, relatively stable material, a lower motion resolution is usually sufficient.
Deflicker — When enabled, small-scale illumination differences between nearby frames are corrected before temporal fixes are applied. Enable this for material with residual flicker that has not already been addressed upstream, to prevent the fix from introducing tonal inconsistencies at repaired pixels.
Constant — When selected, all parameters are held constant across the entire clip. Disable this for material where the level of dirt changes significantly over time, allowing parameters to be keyframed via F-Curves for different sections of the clip.
When to use another effect
Dustbust does not offer size constraints (minimum/maximum pixel area) or the ability to switch between spatial and temporal fix modes based on mark size. For any of these requirements, use Auto Dirt Fix.
5. Auto Dirt Fix — Overview
Type: Automatic | Effect: Auto Dirt Fix | Complexity: Intermediate to Advanced
Auto Dirt Fix is PFClean's primary tool for systematic dirt and dust removal. Where Dustbust optimises for speed, Auto Dirt Fix provides comprehensive control over every stage of the detection and repair pipeline. It is capable of addressing everything from microscopic dust grains through to large print debris and coloured emulsion damage, provided parameters are tuned appropriately for each defect type.
The effect identifies dirt by examining frame-to-frame differences across a motion-compensated image sequence. Pixels that appear anomalous in the current frame but are consistent across neighbouring frames — allowing for image motion — are classified as defects and submitted to the repair algorithm.
Structure of the effect
Auto Dirt Fix is organised into three functional areas, each covering a distinct stage of the pipeline:
Motion Analysis — Controls governing the accuracy and temporal reach of the motion estimation used for detection and temporal repair.
Detection Parameters — Controls specifying what constitutes a piece of dirt: its polarity, contrast, size range, motion range, and which channels to use.
Fix Parameters — Controls specifying how identified dirt is repaired: whether spatially or temporally, how repair boundaries are blended, and whether grain is applied to the patch.
The Results graph at the bottom of the panel provides a visual history of how much work the effect has done per frame (reported as fixed pixels, percentage of image data, or number of fixed areas), and the Show overlay allows the detection mask to be visualised in real time directly in the Cinema.
6. Auto Dirt Fix — Detection Parameters
Motion Analysis
Before detection can occur, PFClean must estimate how the image is moving between frames so that genuine scene motion is not confused with dirt.
Accuracy — Sets the quality of motion estimation. Low is faster; Normal is the practical default for most materials; High should be reserved for material with rapid or complex motion where lower accuracy produces false positives. Note that increasing accuracy increases processing time proportionally.
Lookahead — The number of frames searched before and after the current frame. The default value of 1 means the previous and next frames are used. Increasing this allows the detection of dirt that persists across multiple frames. A value of 2, for example, will catch marks that last up to two consecutive frames.
Smoothness — Controls how smooth the generated motion field is allowed to be. Higher values produce more regular, generalised motion fields. Lower values allow the field to closely track fine-grained per-pixel motion, at the risk of fitting noise. The default is appropriate for most material; increase it when motion fields look erratic or spiky on review.
Detection Parameters
Type — Specifies whether to detect Dark marks, Light marks, or both. Most airborne dust and sparkle reads as light (bright) against the image. Carbon-based debris and opaque foreign matter more commonly reads as dark. Coloured emulsion chips may appear as either, depending on which dye layers have been removed.
Min Contrast % — The minimum contrast a defect must exhibit against the surrounding background, expressed as a percentage of the full pixel intensity range. The default is 10%. Lowering this value captures faint dust marks; raising it restricts detection to only the most visually prominent damage.
Size (Min / Max) — The minimum and maximum number of pixels a piece of dirt can contain. The default minimum is 10 pixels at 2K resolution. There is no maximum by default. Defining a tight size range is one of the most powerful ways to isolate specific defect types: a small maximum (e.g. 50 pixels) restricts detection to fine dust and sparkle; a large minimum (e.g. 500 pixels) restricts detection to large chunks of print debris. This parameter is essential when running a multi-pass stack (see [Section 9](#multi-pass)).
Motion (Min / Max) — Constrains detection to parts of the image undergoing a specified range of between-frame motion. Setting a low maximum motion value (e.g. 5 pixels) restricts cleaning to near-static areas. Setting a minimum motion value can isolate debris that is physically moving across the frame — a tell-tale sign of gate-level contamination rather than emulsion damage.
Channels — The RGB channels used for detection. Restricting detection to two channels (typically excluding the blue channel on noisy material) while leaving all three channels active for the fix is a standard technique for improving detection accuracy without sacrificing repair completeness.
Use Defect Map — When active, use an infrared defect map stored in the clip's alpha channel rather than frame-difference analysis. See Section 8.
Sample — Draw a rectangle in the Cinema over a representative piece of dirt to have PFClean automatically estimate suitable detection parameters. This is a useful starting point for unfamiliar material, but should always be reviewed and refined manually.
7. Auto Dirt Fix — Fix Parameters
Once detection has identified a set of defect pixels, the Fix Parameters control how those pixels are replaced.
Type: Spatial (Spa.) and Temporal (Tem.) — A temporal fix sources replacement pixels from adjacent frames after motion compensation, producing the most natural result for moving image content. A spatial fix sources replacement pixels from the surrounding area within the same frame, which is faster and does not require motion data, but can introduce visible patches in areas of high texture or fine detail. When both are selected simultaneously, a Threshold value determines which mode is applied to each area: marks above the threshold in size are fixed temporally, marks below it are fixed spatially.
L (Link) button — When enabled, temporal fix areas that do not match the surrounding image well — due to colour balance differences between frames or a failed motion estimate — are automatically downgraded to a spatial fix rather than rejected entirely. This increases fix completeness at the cost of potentially introducing more spatial patches.
D (De-flicker) — When enabled, frame-to-frame illumination differences are corrected during temporal fix application, preventing the repaired pixel block from exhibiting a slight brightness or colour shift relative to its surroundings.
Dilation Size — Expands each detected dirt area by the specified number of pixels before the fix is applied. The default is 1 pixel in each direction. A small dilation ensures the repair covers the full extent of a defect including its soft edges. The G button gangs horizontal and vertical dilation together.
Blend Radius — When enabled, the boundary of each repaired area is expanded and blended into the surrounding image after fixing, softening the transition between patch and background. This is essential for avoiding hard-edged "cutout" artefacts at repair boundaries, particularly on fine-grained film stock where abrupt tonal transitions are immediately visible see also Section 13.
Channels — The RGB channels that are actually repaired. This can be set independently from the detection channels: for example, detecting only on red and green to avoid blue-channel noise, but repairing all three.
Grain/Noise Preset — Renders a grain or noise overlay on top of each repair using a stored grain profile. This is a critical quality step for organic film material. A repair made without grain matching will appear as a smooth, textureless "dead zone" against the surrounding grain field — immediately visible on playback. Match this to the grain profile of the scan, either using a profile derived from a clean area of the same clip or from PFClean's preset library.
Store Fix Results To Disk — When enabled, completed fix data is written to disk as it is generated. Returning to a previously processed frame does not require recalculation. Enable this for long clips or high-volume work to avoid repeated processing overhead.
The Results Graph and Show Overlay
The results graph displays per-frame fix data as a bar chart. Switch between Pix (fixed pixels), % (percentage of frame), and Fix (number of discrete fixed areas) depending on what metric is most useful for your QC pass. The View button overlays the current detection overlay directly in the Cinema in a configurable colour, allowing you to confirm the detection is isolating only genuine defects before committing to a full cache pass.
8. Auto Dirt Fix — Defect Maps
When a film scanner is equipped with an infrared channel, it can generate a pixel-accurate defect map that identifies every physical discontinuity in the emulsion independently of the visible light image. This data arrives in PFClean via the clip's alpha channel and is one of the most reliable inputs available for dirt and dust removal.
Setting up a defect map
1. In the Clip Panel, set the Defect Map parameter to either White Alpha (defect pixels stored as alpha = 1) or Black Alpha (defect pixels stored as alpha = 0), depending on your scanner's convention.
2. Once set, PFClean stores the defect map data and removes it from the alpha channel, making the alpha available for other uses.
3. Use the View and Defect Threshold controls in the Clip Panel to confirm that defect pixels are correctly classified before proceeding.
Using a defect map in Auto Dirt Fix
When a defect map has been configured, adding a new Auto Dirt Fix effect will automatically activate Use Defect Map and default the Fix Mode to Spa. + Tem. (both spatial and temporal). No frame-difference detection analysis is required: the map provides the complete pixel-level mask. This works very similarly to how the defect map works in the Digital Wet Gate, however you now have per clip control and the ability to adjust the way the fixes are applied.
Spatial fixing is appropriate for most small defects and has the significant advantage of requiring no motion analysis, making it substantially faster than a temporal-only pass. For large defects, temporal fixing will generally produce a higher quality result, and the combined Spa. + Tem. mode with an appropriate threshold handles both cases automatically.
The infrared defect map approach is the gold standard for dirt and dust removal on archive material scanned with the capability: it eliminates the detection phase entirely and provides a pre-validated mask that no frame-difference algorithm can match in accuracy. Should you choose not to utilize the Defect Map, PFClean provides the flexibility to revert to its industry-proven native detection algorithms. This process is straightforward: simply deactivate the Use Defect Map option, and Auto Dirt Fix will automatically begin recalculating detection via its own internal algorithm.
9. How PFClean’s dirt removal differs from generative AI
Scanning film with an active infrared channel generates a pixel-accurate defect map that captures the physical state of the emulsion independently of the visible image. This identifies the exact location of physical discontinuities, dust, dirt, and scratches, using measurable, photographic information rather than algorithmic estimation of the scene.
This is where PFClean's approach diverges fundamentally from generative AI-based restoration. Rather than relying on learned image priors to synthesise plausible replacement pixels, PFClean repairs the defects identified by the infrared pass using real image information drawn from the same frame and its neighbouring frames. Damaged areas are reconstructed from genuine photographic detail that already exists within the source material, not from predicted or fabricated content.
For archival restoration, this distinction is critical: the objective is to recover authentic image information wherever possible, not to invent it. By prioritising real frame data over synthetic prediction, PFClean delivers a preservation-focused result that maintains the technical integrity of the scan without introducing details that were never present in the original footage.
For a more detailed discussion of this philosophy, see Remastering Archive Footage for Redistribution: Fidelity, Not Fabrication.
10. Advanced Workflow: Multi-Pass Stack Strategy
Skill level: Intermediate–Advanced
Attempting to force a single Auto Dirt Fix instance to simultaneously address fine dust, large print debris, and coloured emulsion chips is the most common cause of poor results. Each defect type requires fundamentally different detection thresholds, size ranges, and channel configurations. Trying to satisfy all of them with a single parameter set produces compromised detection and unwanted processing artefacts. The solution is to decompose the problem into discrete, targeted layers using multiple Dustbust & Auto Dirt Fix instances stacked in sequence in the Workbench.
How to build the stack
Each effect processes the output of the one above it in the stack. This means that a fine-dust pass at the top of the stack produces a slightly cleaner frame for the large-debris pass below it, reducing the chance of the broader pass re-detecting partially-repaired areas from the finer pass. This works similarly to how we tackled a complex stabilisation in our previous guide.
An example three-layer structure
Pass | Effect Name | Tool | Type | Settings | Fix Method | Blend Radius | Grain Preset |
01 | DB - Fine Dust | Dustbust | Light | Motion: 1/2 Threshold: 5% Dilate: 1px Reject: 67% | – | – | – |
02 | ADF - Moderate Dark Debris | Auto Dirt Fix | Dark | Min Contrast: 2.4% Size: 7–150px | Spatial + Temporal | 1px | Applied |
03 | ADF - Large Dark Damage | Auto Dirt Fix | Dark | Min Contrast: 2.4%+ Size: 100px+ (no max) Channels: Green, Blue | Temporal | 2px | Applied |
Naming effects
PFClean allows each effect in the Workbench stack to be renamed. For multi-pass stacks, this is not optional, it is essential housekeeping. When returning to a project days or weeks later, unnamed stacked instances of the same effect are indistinguishable. Use consistent, descriptive names: `DB - Fine Dust`, `ADF - Moderate Dark Debris`, `ADF - Large Dark Damage`. A clearly labelled stack is a maintainable stack.
Combining Effects for Seamless Results
While Auto Dirt Fix can be calibrated to detect all types of dirt and debris, it is most effective for medium-to-large artifacts where advanced blending controls are essential. For footage containing an underlying layer of fine sparkle, an efficient approach is to first utilise the Dustbust effect to quickly clear away the sparkle. This allows the Auto Dirt Fix to focus on repairing the more problematic defects in the shot. This layered stacking strategy is highly recommended when dealing with heterogeneous damage, such as varying types, sizes, or polarities, or when aggressive parameter adjustments are needed for specific defect categories without introducing artifacts elsewhere.
Combine this approach with caching
Multiple complex Auto Dirt calibrations can be computationally intensive. To optimise your workflow, consider disk caching the lowest Auto Dirt Fix layer in your stack. Any additional effects placed below this layer will then process from the cached data, allowing you to continue working smoothly without repeatedly recalculating the same operations.
For a more in-depth explanation of caching workflows, including this approach and best practices for optimising performance, see our complete technical caching guide found here.
11. Advanced Workflow: Masked Region Isolation for Auto Dirt Fix /Dustbust
Skill level: Advanced
The Multi-Pass Stack Strategy demonstrates how running multiple detection passes with different parameter sets can catch a wider range of dirt and damage than a single pass. Masking extends this same principle spatially: rather than adjusting when a pass runs, you constrain where it runs, allowing a single region of the frame to be calibrated independently of the rest of the image.
This becomes essential in situations where a specific area carries dirt at a contrast level that doesn't suit the settings needed for the rest of the frame. A clear sky is the classic case: light, sparse dirt sits at low contrast against a dense, evenly exposed negative, while the rest of the shot may contain higher-contrast detail that would be falsely flagged if detection parameters were pushed to catch the sky dirt. Raising Min Contrast % or lowering Size thresholds enough to catch the sky artefacts typically over-corrects everything else, and the usual remedy, manually undoing false corrections or hand-cleaning the sky frame by frame, is slow and inconsistent across a shot.
Masking removes this trade-off by letting the sky (or any other problem region) run its own isolated detection pass, with the rest of the frame handled separately at conventional settings.
The workflow
Draw a new Auto Dirt Fix or Dustbust effect and attach a mask covering only the problem region, in this case, the sky matte in the background.
With the mask active, open Detection Parameters and calibrate independently of the rest of the frame. Because detection is now constrained to the masked area, you can raise sensitivity, lowering Min Contrast % or widening the Size range, without risk of the rest of the image being affected.
Refine the mask edge as needed (feather or garbage-matte the boundary) so detection doesn't bleed into adjacent high-detail areas at the horizon line or around foreground elements.
Add a second Auto Dirt Fix or Dustbust pass, unmasked or masked to the inverse region, using conventional settings appropriate for the rest of the frame. This pass can be placed either before or after the masked pass in the stack.
Review the composite result across a representative range of frames, checking that the masked pass isn't catching cloud detail or gradient banding as false positives, particularly in skies with subtle tonal variation.
Apply Grain and Blend Radius per effect as needed, since the two passes may require different repair blending depending on the density and grain structure of each region.
By isolating detection spatially rather than only temporally, masked passes let you push sensitivity hard in the areas that need it, skies, flat walls, water, without inheriting the over-correction cost across the rest of the frame. Combined with the Multi-Pass Stack Strategy, this gives you control over both when and where each detection pass operates, which is often the difference between an automated clean-up and a shot that still needs manual finishing.
12. Advanced Workflow: Channel Isolation for Emulsion Chips
Skill level: Advanced
When physical damage affects the upper layers of a film's emulsion, such as deep gouges, chips, or surface abrasion, the damage cuts through specific dye layers rather than all three simultaneously. Because colour negative and colour print film build their image from three layered dyes (cyan/magenta/yellow, corresponding broadly to the red/green/blue channels), physical damage that removes one or two layers will exhibit a distinct colour bias in the scan.
The most common presentation is sharp blue or green fragments, where the upper dye layer has been removed, leaving the lower layers exposed. Standard RGB composite detection will struggle to isolate these marks cleanly because their contrast against the composite image is lower than their contrast against a single channel.
Our article about Film Fade covers the emulsion layers and on modern motion picture film stocks.
The workflow
1. In the Auto Dirt Fix effect, open the Detection Parameters and set Channels to a single channel — typically Blue or Green, whichever exhibits the strongest colour bias for the specific chip type in your material.
2. Leave the Fix Channels set to RGB so that the repair addresses all three channels once the mask has been defined.
3. Adjust Min Contrast % relative to the isolated channel only. Because the chip has high contrast against that single channel, you can often use a tighter threshold than you would on the RGB composite, dramatically reducing false positives.
4. Set an appropriate Size range to prevent detection spilling into legitimate scene detail. Emulsion chips tend to have sharp edges and compact shapes; constraining the maximum size helps exclude gradual tonal transitions.
5. Enable Blend Radius to soften repair boundaries, as emulsion chip edges are often hard and regular, making unblended patches visually obvious.
6. Apply the appropriate Grain preset to match the repaired area to the grain field of the surrounding emulsion.
This channel-isolation approach dramatically increases the relative contrast of the target artefact against the detection model, producing a clean, accurate mask that composite detection cannot reliably achieve.
13. Advanced Workflow: Blend Radius and Grain Matching
Skill level: Intermediate
Spatial and temporal patches can look technically correct, the defect is removed, the underlying image data is plausible, yet still read as wrong on playback. The most common cause is an insufficiently integrated repair boundary.
Film is an organic medium. Its grain is inherently random and spatially textured. When a repair places a mathematically correct patch over a defect area, the boundary between the patch and the original image creates a zone where grain density, colour noise, and luminance texture change abruptly. Even a fraction-of-a-percent difference in grain presence is detectable by the eye, particularly on a calibrated display or projected print.
Blend Radius
Within the Auto Dirt Fix advanced panel, the Blend Radius control expands the detected area and blends the repair boundary into the surrounding image before the patch is finalised. Think of it as a soft feather applied to the repair mask, rather than a hard edge.
Start with a small increment above the default, a Blend Radius value of 1–2 pixels is sufficient for most fine dust repairs at 2K. For larger debris repairs where the patch covers a significant image area, a radius of 3–5 pixels or more may be appropriate. Avoid excessive values: a blend radius that is too large can pull incorrect pixel colours from outside the defect boundary into the feathered region, creating a visible softening halo.
Grain Matching
The Grain/Noise Preset in the Fix Parameters renders a grain texture over each repaired area. For this to be effective, the preset must be derived from the actual grain character of the material being restored, not from a generic library preset.
Best practice workflow:
1. Before dirt removal, create a grain sample from a clean area of the same clip using PFClean's grain analysis tools (De-Grain effect or grain sampling in Paint). Store this as a named preset.
2. Apply this custom preset in the Auto Dirt Fix Grain/Noise Preset field.
3. After processing, use the Show overlay and single-frame scrubbing to verify that repaired areas are texturally consistent with the surrounding grain field on playback.
If the material has significant variation in grain density between scene-light and scene-dark areas (as is characteristic of finer film stocks), consider whether a single grain preset is sufficient, or whether different grain profiles are needed for repairs in highlights versus shadows.
You can see examples of PFClean matched grain in our article on Understanding Film Grain in Digital Restoration.
14. Advanced Workflow: Correcting Overcorrection with the Q/A Panel
Skill level: Intermediate
Tuning Dustbust or Auto Dirt Fix parameters is inherently a balancing act. On thin, low-contrast, or unevenly exposed material, pushing detection aggressively enough to catch every defect will almost always introduce some overcorrection, legitimate detail mistaken for damage and repaired unnecessarily. Backing the settings off to protect the image means defects slip through and have to be handled manually. Historically, this has meant a trade-off: restoration artists either accept some visible damage and correct it manually or accept some loss of genuine image content.
The Q/A panel removes this trade-off. Rather than adjusting effect parameters and re-running detection, the Q/A panel lets you view the defect mask directly and selectively undo individual fixes, frame by frame or across a range, without touching the underlying Auto Dirt Fix or Dustbust settings. This is a targeted correction tool, not a blanket undo, it operates on the mask itself, so you can remove a single erroneous repair while leaving every correct detection in the pass untouched.
This means detection can be tuned aggressively, maximising the defects caught automatically, with the confidence that any resulting overcorrection can be identified and reversed quickly afterwards, rather than accepted as the cost of a high catch rate.
The workflow
Run your Dustbust or Auto Dirt Fix pass with detection parameters set aggressively enough to catch the full range of defects in the shot, accepting that some overcorrection is likely.
Open the Q/A panel to review the defect mask generated by the pass.
Step through the affected frames and identify any areas where genuine image detail has been caught and repaired in error.
Use the panel's tools to undo the fix in those specific areas, on a single frame or across a frame range, without altering the effect's detection settings.
Leave all correctly identified repairs in place, the correction is isolated to the overcorrected regions only.
Re-check the shot at full resolution to confirm the balance between defect removal and preserved image detail.
Used this way, the Q/A panel lets you set detection thresholds for maximum catch rate rather than for safety, since any overcorrection it introduces is quick to isolate and reverse, giving you both aggressive automated cleaning and precise manual control over the same pass.
15. Troubleshooting Common Issues
Detection is missing visible dirt
Cause: Min Contrast % is set too high, or the Type setting does not match the polarity of the damage.
Solutions:
- Reduce Min Contrast % incrementally and use the Show overlay to verify detection coverage.
- Confirm that the Type setting (Dark / Light / Both) matches the actual polarity of the dirt visible in the Cinema. Bright sparkle requires Light; opaque debris typically requires Dark.
- If damage is a specific colour rather than grey, try isolating the detection to the most affected channel see Section 12
Detection is removing clean image detail
Cause: Min Contrast % is too low, or the Size range is too broad, or there is significant motion in the scene that is being misread as dirt.
Solutions:
- Increase Min Contrast % until the false detections disappear from the Show overlay.
- Set a maximum Size value to prevent large areas of legitimate motion from being classified as defects.
- Increase Motion Analysis Accuracy.
- Increase Reject % to filter out low-confidence detections.
- Use the Motion parameter range to restrict detection to areas undergoing a specific range of motion.
Repairs are visible as "flat" or "dead" patches
Cause: No grain preset applied, or Blend Radius is too low.
Solutions:
- Apply a grain or noise preset matched to the grain profile of the material, see Section 11.
- Increase Blend Radius to soften repair boundaries.
- If the repair is sourced temporally and the colour balance between frames is inconsistent, enable the D (De-flicker) button to compensate.
Large debris is being missed while fine dust is being caught, or vice versa
Cause: A single effect is trying to address defects at opposite ends of the size spectrum.
Solution:
Implement a multi-pass stack see Section 10. Target each instance to a specific size range and calibrate Contrast and Type independently for each layer.
Temporal fixes look offset or misaligned
Cause: Motion estimation has failed for that area, or the motion between the reference frames is too large for the current Lookahead value.
Solutions:
- Increase Motion Analysis Accuracy to Normal or High.
- Increase Lookahead to give the motion solver more reference frames.
- For areas where temporal fixing cannot reliably reconstruct the image, switch to Spatial fix mode for that size range, or use the Manual Dirt/Dust Fix tool to address those frames individually.
Defect map is not being detected correctly
Cause: The Defect Map parameter in the Clip Panel is set to the wrong polarity (White Alpha vs. Black Alpha), or the Defect Threshold is not calibrated correctly.
Solution:
- Examine the alpha channel of the clip using the RGBA controls in the Cinema (press ⌥/Alt - A).
- Confirm whether defect pixels appear as white or black in the alpha channel, then set the Defect Map parameter accordingly.
- Adjust the Defect Threshold to ensure all visible defects in the alpha channel are classified correctly before processing.
Processing speed is too slow for the volume of material
Cause: Motion Analysis Accuracy is set to High, or Store Fix Results To Disk is disabled on a long clip.
Solutions:
- Reduce Motion Analysis Accuracy to Normal or Low, particularly for material with slow or low-complexity motion.
- Enable Store Fix Results To Disk to prevent recalculation when returning to previously processed frames.
- For material where fine-grained control is not required, switch from Auto Dirt Fix to Dustbust, which is optimised specifically for throughput.
16. Automatic Film Dirt & Dust Removal: Quick Reference
This decision tree guides you through selecting the right PFClean workflow for your material.
Step 1: Is stabilisation complete?
No — Return to the top of the Workbench stack and address stabilisation first. Unstable frames will degrade detection accuracy.
Yes — Proceed to Step 2.
Step 2: Does the scan have an infrared defect channel?
Yes — Configure the Defect Map in the Clip Panel, then use Auto Dirt Fix with Use Defect Map active. This is the most accurate route available.
No — Proceed to Step 3.
Step 3: Assess the damage type and volume
- Is the damage limited to fine, fairly uniform, single-frame dust and sparkle, and is processing speed important?
→ Use Dustbust. Tune Type, Thresh %, Detect channels, and Reject % to the material.
- Is the damage varied — multiple sizes, types, or colours — or does it require precise size/channel control?
→ Use Auto Dirt Fix. Proceed to Step 4.
Step 4: Is the damage homogeneous or heterogeneous?
- Is all the visible damage broadly the same size and character?
→ Use a single Auto Dirt Fix instance, tuned to that damage type.
- Is the damage mixed — fine dust plus large debris, or white sparkle plus coloured emulsion chips?
→ Build a Multi-Pass Stack see Section 10. One targeted effect per defect category.
Step 5: Does the damage show a strong colour bias?
- Yes — Use channel isolation in Detection Channels see Section 10 to detect on the biased channel only.
- No — Use RGB detection.
Step 6: Quality check
- Are repair boundaries visible on playback?
→ Increase Blend Radius and apply a matched Grain/Noise Preset see Section 13.
- Are repairs offset or misaligned?
→ Increase Motion Analysis Accuracy or Lookahead; consider switching to Spatial fix mode for affected frames.
- Is detection catching clean image detail?
→ Increase Min Contrast %, add a Size maximum, or increase Reject %.
FAQ
Should dirt removal come before or after stabilisation?
After. Dirt detection compares frames, so gate weave or jitter makes stationary detail look like moving dirt, stabilise and de-flicker first, then clean, then finish colour and grain when required.
What’s the difference between Dustbust and Auto Dirt Fix?
Dustbust is a high-speed pass for fine, uniform dust; Auto Dirt Fix adds full control over size ranges, channels, and spatial-versus-temporal repair for complex or mixed damage.
What is an infrared defect map?
A pixel-accurate mask of physical emulsion defects captured by scanners with an IR channel; PFClean can use it directly, replacing detection with measured data, the most accurate route available.
Why do some repairs look like smooth, flat patches?
The repair lacks grain matching or boundary blending; apply a grain preset sampled from the clip and enable Blend Radius so the patch integrates with the surrounding grain field.
How is this different from AI dirt removal?
PFClean reconstructs damaged areas from real image data in the same and neighbouring frames; generative AI predicts plausible pixels from learned priors, which can introduce detail that was never photographed.
See PFClean's Dirt & Dust Removal in Action
Experience these workflows on your own footage. Book a demo with our product specialists and work through your material in real time, or upload a clip and we'll create a custom before-and-after demonstration tailored to your project.
From one-click Dustbust passes to precision multi-layer stacks, discover how PFClean delivers professional-grade dirt and dust removal on even the most challenging archive material.
Related Articles
About the Author
Adam Hawkes is a PFClean Product Specialist and restoration expert with over 20 years of hands-on experience in film and video restoration. Trained in film handling and film camera operation, Adam has contributed to more than 100 productions, including some of cinema's most celebrated titles. His expertise combines deep technical knowledge of restoration workflows with practical understanding of the physical and optical characteristics of film.