Grain - Guillermo Algora - Visual Effects Compositor

Guillermo Algora
Go to content
GRAIN:

Very small random fluctuations that are visual artefacts or defects in the image. This is a side effect, mostly from the image capture process. Digital Camera noise is electrical interference. Analogue film grain is grains of silver. Digital grain is more accurately described as noise but we will refer to both as grain.

Primary Properties:

Intesity:

Intensity is the amount of negative and positive multiplication of the image by the noise. This results in the strength and color of the grain. The intensity is mainly dependent on:

  • The luminance value of the image.
  • The ISO/ASA of the image.

Film grain is generally more color neutral than in digital images. Digital grain has more color "fizzle" between the pixels.
Digital has less variation between the overall color between channels.
Film has the most intensity in hte bue channel.

Size:

Grain pixels clump together to generate grain of a particular size.
  • High frequenzy / small: usually no more than a pixel.
  • Medium frequency.
  • Low frequency: this can be large blobs of changing texture.

Film / Digital differences:

  • With both film and digital, the higher the ISO/ASA the larger the grain.
  • Film: the size of grain can change according to the luminance of the image, with darker areas of the image having larger grain.
  • Digital grain generally stays roughly the same size within a single RGB channel.

Secondary Properties:

Pattern Irregularity:

Higher ISO/ASA results in more, with in Film looking more clumpy due to larger chunks of silver. Film will always have a more clumpier / irregular grain look to it. Some digital cameras produce exactly the same pattern in each RGB channel, but with different intensity, and so require the same seed for their simulation.

Some digital noise can have filtering artifacts:
  • Banding.
  • Dot arrays.
  • Patterning often, with clumps of noise grouping into little vertical or horizontal shaped clumps.

Softness:

Grain is a primarily gaussian like noise pattern with exposure differences. Sometimes to match the plate grain a secondary gaussian blur or sharpen is required.

Pipeline Considerations:

  • Cropped or transformed scans: then any grain simulations need to have those transforms applied to the grain before being used to match.
  • Different exposures: if the scan has had an external client grade or internal neutral grade applied to it then this will affect the grain and has to be taken into account to get a perfect match. Also if the scan is exposed differently in camera (colour temperature or intensity) then it will have a different grainr response to the surrounding shots in the cut. Grading through a neutral grade is a useful strategy to make sure the grain is consistent between shots.
  • Negative values in Linear: this breaks many degrain and regrain nodes.Once choice here is to add to the minimum floor so it is positive and then substract the same amount after. Alternatively color transforming the scan back into its original Log space with a lut will ensure no negative values as well as more subtle control between contrast areas of the grain.
  • Request grain set from your lead, before doing your own, to keep consistency and be an efficient team.
  • Do not use a 3D viewer lut when adjusting grain, as the channels are "mixed" together. 1D lut is fine (sRGB, rec.709 ), to be safe you can use 'Raw / None' and check it with the show viewer lut to finish.
  • Only have the zoom settings in the viewer set to divisions of a factor of 2, i.e. 100%, 200%, 400%, etc. to avoid seeing pixe;s
  • Examine each RGB channel separately first and then lastly together.
  • Gamma slam and exposure to create contrast makes easier to judge intensity and size.
  • Always compare back to the plate. Rocking back and forth between frames helps size quickly.
  • Use a QC tool or high-pass difference filter to extract the grain and compare, however this sometimes lead to artifacts and it's not 100% accurate. It is helpful to identify mismatches quickly.
  • Don't regrain before lens distortion or other lens artifacts.
  • Hard edges for grain better, as soft edges duplicate grain and softness it.

Regrain Methods:

Extracted:
Grain scattered procedurally, derived directly from a scan. This technique may bring back a ghost image of the original grain plate.

Simulated:
Offers the most flexibility. Simulated grain is a digitally gaussian noise pattern created by analysing a scan.

Templating Grain Plates / Simulations:

A single template can be used across multiple shots but it will be specific to:
  • Camera model / Film stock.
  • ISO.
  • Scan crops / Transforms.
  • Neutral Grade for the sequence (only if using them).
  • Heat temperature (only for some cameras / stocks).
  • Interior or Exterior (color temperature).

Grain Tools:

Native Grain Node:
  • Does not adjust the intensity of the grain to match the response curve of the grain.
  • Often multiple grain nodes are applied through keys of different luminance areas of the image to give complete control of size and intensity over the response curve.

F_Regrain:
  • Will automatically try and detect grain size and the intensity response curve. Need to be adjusted manually with its controls. It is slow and require NukeX licence.

DasGrain:
  • Advanced Grain Extraction Node.
  • Attempts to normalize luminance differences and break up the image with organic tilling.
Back to content