Your AI clips come out soft. Every upscaler either re-diffuses every frame for an hour, or smears the video with a blurry resize. RelayVSR does neither: a big model redraws one keyframe in 15, a small decoder carries the detail across the rest — so a whole clip finishes in a streaming pass. ComfyUI shipped it natively on 5 November 2026, and the official weights passed 32,000 downloads in 3 days.

WHAT’S INSIDE
• The ready 2x finishing preset (ComfyUI workflow): load your clip, run once, get a 2x video with the original audio — saved as ProRes .mov, straight into your editor
• The ready 4x preset: same graph, 4x weight pair already swapped in, for real 1080p → 4K delivery
• The free official Comfy-Org 2x template as the study base, unchanged
• A settings library with 12 source-type presets: H3 clips, Seedance takes, phone footage, old video, anime, talking heads, fast action, screen recordings — which scale, which detail dial, what to watch for
• Quick-start guide (EN) + full workflow doc (AR): first upscale in 3 stages, the 5-second test trick, settings table, troubleshooting

No prompt to write, no text encoder, no custom nodes, no API key. Two free model files (Apache-2.0, 4.81 GB + 122 MB for 2x), one diffusion step at CFG 1.0, ~14 GB VRAM at 1080p — Colab T4 or Comfy Cloud if your laptop has no GPU.