Home Creators Posts Import Register Favorites Logout

Downloads

Content

Video : https://youtu.be/Urdt5ibVijM

Hey everyone. New video is up, and this one fixes a problem every AI video creator runs into: your clips look great in the preview, then fall apart the moment you try to upscale them.

The video is about RelayVSR, a video super-resolution model from USTC that just got native ComfyUI support. It takes a different approach than running one giant model on every frame. A big generative model paints a few keyframes, then a small decoder fills in everything between. That relay design keeps it fast (the paper shows almost 30 fps at 1080p) and light on VRAM, and it was built to handle AI-generated mush, not just real-world camera noise.

I tested it on MiniMax H3 output and older renders, and honestly it surprised me. Faces hold together frame to frame, which is where most upscalers fall over.

The freebie RelayVSR upscale workflow from the video is attached to this post.

All model links are in the YouTube description.

For Patreon Supporters check out the additional content member post , https://www.patreon.com/aifuturetech/posts/h3-pika-realvsr-174675193

— Benji

Resources:

RelayVSR: Large–Small Model Collaboration

for Efficient Real-World Video Super-Resolution

https://kopperx.github.io/RelayVSR/

https://github.com/kopperx/RelayVSR

https://huggingface.co/kopper/RelayVSR

ComfyUI Support RelayVSR Github

https://github.com/Comfy-Org/ComfyUI/pull/17511

https://huggingface.co/Comfy-Org/RelayVSR/

Files

RelayVSR in ComfyUI - A Fast Upscale AI Video

Upscale AI video with RelayVSR in ComfyUI — the video super-resolution model that just landed natively in ComfyUI. One-step Wan2.2 5B keyframes plus a Flash decoder, tested on real MiniMax H3 output. This tutorial covers RelayVSR, the real-world video super-resolution model from USTC that got native ComfyUI support on November 5, 2026. RelayVSR takes a different approach than running one giant model on every frame: a one-step Wan2.2 5B diffusion model paints a keyframe every fifteen frames, then a small Flash decoder fills in everything between, streaming off the keyframe latents. I walk through the x2 and x4 models, the two ComfyUI nodes (RelayVSRConditioning and VSRDecode), the detail gain knob for over-sharpening, and how I chained RelayVSR into my upscale stack with DLSS5 enhancement, RTX frame generation and HDR. This is for ComfyUI users and AI video creators whose generated clips come out soft. If you generate with MiniMax H3, AnimateDiff or anything similar, or you just want to rescue old low-resolution footage, this works on consumer GPUs. The paper numbers put it around 14 GB of VRAM at 1080p, and the workflow needs no text encoder. Why this matters: RelayVSR was evaluated on real-world and AI-generated degradation, not just clean synthetic benchmarks. In testing it runs fast (the paper shows almost 30 fps at 1080p versus 7.8 for FlashVSR), holds faces together across frames, and ships as Apache-2.0 with x2 and x4 variants. Native support means two nodes, no workarounds. The free upscale workflow from this video is on my Patreon, and the supporter post has the full MiniMax H3 to 2K pipeline. Resources: RelayVSR project page: https://kopperx.github.io/RelayVSR/ RelayVSR model files (ComfyUI repack): https://huggingface.co/Comfy-Org/RelayVSR RelayVSR original repo: https://huggingface.co/kopper/RelayVSR ComfyUI native support (PR 17511): https://github.com/Comfy-Org/ComfyUI/pull/17511 Freebie RelayVSR upscale workflow : https://www.patreon.com/aifuturetech/posts/174673261 Calliope - Local-first Agentic Idea-to-video Studio https://github.com/benjiyaya/Calliope Chapters: 00:00 - Introduction & Relay VSSR overview 01:52 - Downloading models & file placement 03:06 - Setting up native nodes in ComfyUI 04:47 - Sampler & conditioning setup 05:32 - Adding RTX frame interpolation & DLSS 06:30 - Workflow demonstration: MiniMax H3 upscale 07:54 - Side-by-side comparison & quality results 10:56 - End-to-end generation & upscale pipeline 12:28 - Additional video comparisons 13:59 - Demo showcases & outro If You Like tutorial like this, You Can Support Our Work In Patreon: https://www.patreon.com/c/aifuturetech #RelayVSR #ComfyUI #AIVideo #VideoUpscaling #MiniMaxH3

Comments

No comments found for this post.