Putting back the texture the encoder removed

Why grain works so well here

Every clip went through the same lossy encoder, which removed fine texture and returned a smooth, denoised surface. That smoothness is one of the strongest cues that something is synthetic: stronger than colour, stronger than motion, and largely unconscious.

Real captured images have noise. Film has grain, digital sensors have shot noise, and both are irregular, high-frequency, and present everywhere.

Adding grain does two things at once:

1. It replaces a missing property, so the image reads as captured rather than rendered.
2. It gives every clip the same surface. Thirty generations with one grain layer over them share a texture they did not have individually, which is a form of matching that colour work cannot achieve.

One pass, over everything

The rule that matters: apply grain once, to the finished timeline, on an adjustment layer above everything.

Applying it per clip gives you thirty grains with slightly different characteristics, and the difference is visible at every cut, which is worse than no grain at all, because now the cuts have an extra thing changing across them.

In Resolve: an adjustment clip on the top track spanning the whole timeline, with Film Grain on it. In Kdenlive or Shotcut: a grain filter on a track above, or a grain video file on a track above set to overlay or soft light blend at low opacity. In CapCut: a grain overlay applied to the whole project.

Getting the amount right

The two things people get wrong are amount and size.

Amount. You should not be able to see the grain when watching normally. You should see it if you pause and look. If the grain is noticeable in motion, halve it. A useful test: toggle the layer on and off: the footage should look slightly wrong with it off and unremarkable with it on.

Size. Grain has a spatial scale, and that scale must relate to the output resolution. Grain sized for 4K applied to a 1080p export looks like fine noise; grain sized for 1080p and then upscaled looks like blotches. Set it for the resolution you are delivering at, and if you change delivery resolution, revisit it.

There is also a delivery consideration. Grain is high-frequency detail, which is exactly what video compression removes first, so heavy grain plus a low-bitrate platform re-encode produces a smeary mess. Keep the grain light and the bitrate adequate, which the export lesson deals with.

Halation and bloom

Real lenses and real film do something else generated footage does not: bright areas spill slightly into their surroundings. Film halation gives a warm red glow around highlights; lens bloom gives a general soft spread.

A small amount of this sells a shot as photographed more than most people expect. In Resolve, the Glow effect at low intensity restricted to highlights. In any editor, duplicate the clip, blur the copy heavily, restrict it to bright areas with a luminance key, and blend it back at low opacity.

The same warning as grain: subtle. Visible bloom is a filter.

Defocus and vignette

Two more properties of real optics worth adding deliberately rather than asking the model for.

Defocus. A keyframed blur, masked, gives you the rack focus that the model performs badly. It is deterministic and adjustable, which the generated version is not.

Vignette. Real lenses darken slightly toward the corners. A very light vignette focuses attention and reads as lens character. A heavy one reads as an Instagram filter from 2012.

Denoise before regrain, sometimes

A counter-intuitive professional move worth knowing. If a generated clip has visible compression artefacts or inconsistent noise of its own, cleaning it first and then adding uniform grain produces a better result than adding grain on top of a mess.

Resolve’s free tier does not include the full noise reduction tools, which is one of its real limitations. The free route is FFmpeg’s hqdn3d or nlmeans filters:

ffmpeg -i in.mp4 -vf hqdn3d=1.5:1.5:6:6 -c:v libx264 -crf 18 clean.mp4

Then grain the clean version in the edit. This is worth doing only when a clip has a noise problem, not as a default.

Today: put a single grain layer over your whole timeline, set it until you can only see it on a paused frame, and toggle it on and off while watching.