What problem does it solve?
Upscaling video with per-frame image upscalers causes temporal flicker and magnifies compression artifacts, and choosing the wrong node pack, model variant, or parameter (like SeedVR2's 4n+1 batch size) leads to OOM errors or degraded output.
Core Features & Use Cases
- Temporal restore pipelines: Builds the downscale-first → SeedVR2 (or FlashVSR) → RIFE interpolation → VHS encode graph, with exact node classes, model files, and VRAM tiers from 8 GB to 24 GB+.
- Quick local path: Uses already-installed ESRGAN models (e.g. 4x_foolhardy_Remacri) with a 4x→2x supersample plus the built-in ComfyUI 0.26 FrameInterpolate node when no multi-GB download is wanted.
- Use Case: A user has a 720p AI-generated clip at 24 fps and wants a clean 1080p 48 fps result; the skill wires LoadVideo → 0.5x downscale → SeedVR2 3B fp8 → RIFE rife49 multiplier 2 → VHS_VideoCombine with audio passthrough.
Quick Start
Upscale this video to 1080p and double its frame rate using the SeedVR2 and RIFE pipeline in ComfyUI.