Film Grain in Compressed Video
Film grain is fine varying picture texture that can be preserved, simplified, or modeled during video encoding.
A still wall can be difficult to compress when its scanned film texture changes in every picture. The wall is stable, but the grain is additional varying information the encoder has to represent.
Film grain can be part of the intended appearance of a film scan or a deliberately added effect. Its strength, scale and behavior belong to that presentation.
A static scene can have a changing residual
Take a simplified underlying scene sample of 100:
| Picture | Grain contribution | Observed sample |
|---|---|---|
| A | −3 | 97 |
| B | +4 | 104 |
Predict B from A without any movement and the residual is 104 −97 =7. The scene's underlying value did not change; the grain contribution did.
Across many samples, that varying texture weakens the usefulness of directly reusing the previous picture. Finer grain can leave many changing coefficients rather than a small predictable scene correction.
Original scalar illustration, not a film-grain generator or measured codec result.
Direct coding spends data on the visible pattern
An encoder can represent the grain as part of the picture samples. Its predictions, transforms and quantization then handle both the underlying scene and the texture.
At a tight budget, some grain detail can become smoother or form coarser patches. If the coding decisions change sharply between regions or pictures, the retained texture can also appear uneven: one area has stronger grain, then another frame suppresses it.
This can make a film seem to pulse or “breathe,” even when its overall resolution and average bitrate remain unchanged.
A grain tuning can prioritize consistency
x265's documented tune grain emphasizes avoiding artifacts caused by uneven grain distribution. It restricts selected QP-varying behaviors and biases decisions toward more high-frequency content. Its specialized rate control also tries to limit quantizer fluctuations.
That is a specific allocation priority. It is not a promise to reproduce every original grain sample or simply a universal instruction to keep all noise.
A whole-scene review should therefore assess whether the texture remains coherent in motion, alongside the resulting data requirement.
Model the appearance instead
AV1 can encode a cleaner picture and signal parameters for adding modeled grain during output processing.
| Parameter role | What it contributes to the generated appearance |
|---|---|
| Strength versus brightness | Grain can behave differently in dark and light areas |
| Relationships between nearby samples | Gives the texture structure rather than independent noise alone |
| Generation seed | Specifies the generator's repeatable pattern behavior |
The decoder applies the specified synthesis to the output. The reconstructed scene retained for future prediction is separate from that added grain realization.
This can reduce the need to transmit every grain sample in the coded picture while retaining a designed grain-like appearance. It preserves an appearance model, rather than the exact source texture array.
Evaluate texture rather than a single paused speck
Compare passages at the intended scale and in motion. Look at grain strength, clumping, continuity across regions and stability around picture changes. A paused enlargement can emphasize isolated samples without showing temporal pumping.
Pixel-error metrics can also react to a different grain pattern even if its intended texture is similar. A synthesized realization and the original scan need not match sample for sample. Decide whether the comparison aims at exact sample preservation or the intended viewing appearance.
For an Emby viewing version, that distinction explains whether a grain change came from simplification, filtering, inconsistent direct coding, or a different modeled texture. Those are different operations on the film's appearance.