Directors pay millions to preserve film grain through digital pipelines — then audiences, asked which version they like, pick the clean one. Both sides are right. The thesis from perception research: grain’s value isn’t pictorial quality but meaning: texture, time-stamp, and the honest signal of a physical process — which is exactly why viewers rate it lower and artists rate it higher.

What preference studies find

When viewers compare identical images with varying grain, preference falls almost monotonically with noise: clean wins, light grain tolerates, heavy grain loses (Guo et al., 2024). Film-type grain beats Gaussian noise only marginally, and video viewers prefer noise-free outright. Classic granularity research (Lisson, RIT thesis) shows quality judgments track grain most in smooth areas — skies, skin — right where photographers notice it too.

Yet creators keep grain deliberately. A quality-modeling review (Alexandrova, on creative intent) frames grain as texture whose statistics the visual system reads as richness, and codec research (Zeng et al., 2022; IBC, 2023) treats grain preservation as preserving intent — streamers now synthesize grain at the receiver rather than spend bits transmitting it. Grain says: a physical thing happened here, in a time with texture.

The counterpoint: nostalgia or noise?

Skeptics note the emperor’s-new-clothes risk: younger viewers raised on clean digital may read grain as defect, not poetry — preference gaps between creators and audiences suggest exactly that. Heavy grain destroys the shadow detail and color fidelity the photographer worked to capture. And digital “grain” overlays often mimic the look without the structure (grain that ignores exposure and density looks fake to trained eyes).

Takeaway

Shoot grain with intent, not default: fast film pushed for gritty street work, slow film clean for luminous landscapes — and match grain to output size (what sings at 8×10 shouts at billboard). Grain isn’t quality. It’s voice — and like any voice, it should say something.