AV1@Scale: Film Grain Synthesis, The Awakening

Perception of Grain & “Realism”

  • Several commenters dispute the article’s “grain = realism” claim: eyes don’t see grain in normal conditions, and grain obscures scene detail.
  • Others argue our eyes do experience noise, especially in low light, and that added grain can:
    • Increase perceived sharpness and detail.
    • Provide “high‑frequency energy” that compression/optics tend to wash out.
    • Act like visual dithering, hiding banding and compression artifacts.
  • Some distinguish “real” film grain (linked to film crystals and exposure) from generic RGB noise; the latter looks artificial and ugly.

Cultural & Aesthetic Conditioning (24fps, nostalgia)

  • Many see grain and 24fps as artifacts of old technology that became aesthetic norms purely through familiarity and association with “cinema.”
  • Debate over whether higher frame rates should replace 24fps:
    • One side: 24fps is an arbitrary cost‑saving compromise; higher FPS objectively improves motion, especially for action.
    • Other side: a century of 24fps work makes it culturally loaded; changing it meaningfully alters the “cinematic” feel and will take generations.
  • Parallel examples: vinyl “warmth,” tube amps, CRT blur, film jitter, window muntins, vignetting, shallow depth‑of‑field “blurry vignette” looks.

What Netflix’s AV1 Film Grain Synthesis Is Doing

  • Core idea: denoise the master, compress the cleaner image, then reconstruct grain on decode using AV1’s Film Grain Synthesis (FGS) tools.
  • Rationale:
    • Encoding literal noise wastes bits or smears it over large areas, reducing sharpness of actual edges and textures.
    • Removing noise first makes video more compressible; saved bits can preserve more scene detail at a given bitrate.
  • Some note AV1 FGS has existed but was hard to tune; Netflix’s story is about automating it “at scale” with adaptive variants.

Skepticism & Fidelity Concerns

  • Multiple commenters think Netflix’s example looks overly blurred, with re‑added grain that resembles generic RGB noise, not true film grain.
  • Concern: grain (and its temporal behavior) can act as dithering and encode fine detail over time; aggressive denoising then adding fake grain loses that detail.
  • Others counter that:
    • Noise itself doesn’t contain signal; denoisers may discard some true detail, but FGS still beats encoding raw noisy frames at the same bitrate.
    • Still‑frame comparisons understate motion effects, but streaming constraints make some lossy approach unavoidable.

Creative Intent, User Control & Physical Media

  • Some insist grain decisions belong to filmmakers in post, not streaming engineers; others argue client‑side grain is a sensible bandwidth optimization and should be user‑toggleable.
  • A subset of commenters reject all of this as “stepped‑on product,” wishing for lossless or physical media instead, though others point out the impracticality of uncompressed 4K+ video sizes.
  • Overall split: some love grain (especially for older or 16mm‑style content); others want it gone, viewing it as obsolete noise rather than essential texture.