Denoise and Regrain for VFX
Follow along with the tutorial footage by joining the cutting club and support the free youtube tutorials today!
Invisible Object Removal: Restoring Original Noise in DaVinci Resolve Fusion
When you are asked to remove an object, paint something out, or perform general image cleanup, the visual edit often stands out—not because the painting is bad, but because the texture is wrong. When you paint on a frame, you lose the original digital noise and grain, creating an unnaturally smooth patch.
This tutorial breaks down a professional workflow used by finishing artists to denoise footage, remove an object, and then surgically re-apply the original offset grain for a seamless result.
The Core Concept
The goal is to separate the image detail from the noise profile. This allows us to clean the image on a smooth canvas and then layer the original noise back on top.
The workflow involves:
Denoising the original footage.
Linearizing the color space for accurate math.
Removing the object (Cleanup).
Extracting the noise using subtraction.
Offsetting the noise to remove "ghosting."
Adding the noise back to the clean plate.
Step 1: Denoising the Footage
All footage contains noise. To get a clean slate for object removal, start by applying noise reduction.
The Tool: Use the standard Noise Reduction tool built into DaVinci Resolve (Fusion page).
Settings: You can use "Faster" mode to keep the tutorial moving, though it may destroy some fine detail.
Performance Tip: If you want to use a higher quality mode but maintain performance, right-click the node and select Cache to Disk > Pre-render.
Step 2: Linearizing for "Channel Booleans" Math
To extract noise accurately, we need to perform mathematical operations (subtraction and addition) on the pixel data. This works best in a linear color space.
Remove Gamma: Use the Gamut tool after your noise reduction. If your footage is Rec.709, set the source to Rec.709 and ensure the "Remove Gamma" checkbox is selected.
Create Two Streams: You need two linear versions:
A linear version of the Denoised footage.
A linear version of the Original (grainy) footage.
Viewing: To view linear footage correctly on your monitor, verify the "View LUT" (the waffle button) is enabled.
Step 3: Image Cleanup (The Patch Replacer)
Now that you have a clean, denoised, linear image, you can remove the object. In this example, we are removing an Apple logo from a laptop.
The Tool: The Patch Replacer is excellent for this task.
Method: Set the fill-in method to Blended Clone.
Execution: Point the source circle to a clean area of the laptop lid and the destination circle over the logo. Match them as close as possible regarding lighting.
Note: At this stage, the logo is gone, but the area is perfectly smooth and lacks the texture of the rest of the video.
Step 4: Extracting the Noise Profile
We need to isolate the noise so we can put it back later. We do this using Channel Booleans to perform a subtraction.
Node: Add a Channel Booleans node.
Operation: Set the operation to Subtract.
The Formula: Input the Original Linear Footage (Background) and subtract the Denoised Linear Footage (Foreground/Green Input).
Concept: Original (Detail + Noise) minus Denoised (Detail) equals Noise.
If you view this node and gamma up, you will see the isolated noise dancing around on a grey background.
Step 5: The "Front Source" Offset Technique
If you look closely at your extracted noise, you will see a "ghost" of the object you removed (e.g., the Apple logo pattern is visible in the grain) because the noise reduction wasn't perfect. If we simply add this back, the logo will reappear as a ghostly texture.
To fix this, we use a technique similar to "Front Source" in Flame to offset the grain.
The Merge Trick: Add a Merge node after your extraction. Feed the extracted noise into both the Background (Yellow) and Foreground (Green) inputs.
Masking: Use a B-Spline tool to draw a shape around the "ghost" logo in the noise. Connect this to the Mask input of the Merge node.
Transforming: In the Merge node settings, use the Center (transform) controls. Because of the mask, this only moves the "Foreground" input inside the masked area.
The Result: Slide the texture until you replace the "ghost logo" noise with clean noise from a nearby area.
Now, the noise profile is consistent and contains no trace of the removed object.
Step 6: Re-applying the Noise
Now we combine our cleaned image with our cleaned noise.
Node: Add a second Channel Booleans node.
Operation: Set the operation to Add.
Inputs: Take your Cleaned/Patched Image and add the Offset Noise signal to it.
Step 7: Final Output
Before sending the footage back to the Color page or rendering, you must reverse the linearization process.
Copy your original Gamut node and paste it at the end of the chain.
Change the setting from "Remove Gamma" to Add Gamma.
The result is a clip where the object is removed, but the film grain/digital noise remains perfectly consistent across the entire frame, standing up to scrutiny even if contrast is added later.
