A more manual, classic approach involving downscaling, removing the blur, and then using "Super Resolution" filters to upscale the video back to 1080p, as described in Infognition tutorials The Role of High-Definition (1080p)
# 2️⃣ Second pass (final file) ffmpeg -i "MosaicFSDSS617 - Natsu Igarashi.mkv" \ -c:v libx264 -b:v 3000k -pass 2 -preset slow \ -vf "scale=1280:-2" \ -c:a aac -b:a 128k \ -movflags +faststart \ "MosaicFSDSS617_Natsu_Igarashi_720p.mp4" reducing mosaicfsdss617 natsu igarashi 1080p
He worked for hours, his eyes burning. The file size was massive. 1080p meant over two million pixels per frame, and a significant percentage of them were currently noise. A more manual