AI video polishing has no standard workflow: downscale, add noise, upscale may be the fix

robertwellesley · reddit · 2026-09-18

A Reddit user shares an exploration of final polishing steps for AI video: there are almost no prebuilt open-source workflows, and output quality mostly depends on the pipeline itself.

He built a rudimentary workflow overnight — downscaling, adding noise, then upscaling — which helped clean things up. His final output is 12/16-bit 1080p, fairly clean but a bit "dusty," produced inside a Singularity-based workflow he has been refining. He asks the community whether downscale+noise+upscale is the only approach, and what nodes, techniques, or models others recommend.

Original post →

More from Multimodal

Multimodal channel →