NVIDIA one-step refiner that upscales low-resolution generated video to up to 4K across different base video generators.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
SoL-Refiner (Speed-of-Light Refiner) is a video refinement model from NVIDIA Labs, developed by the Sana team. It takes low-resolution output from a base video generator and produces a high-resolution video (up to 4K) using a single target-resolution denoising step, avoiding the second multi-step sampling bottleneck of conventional refiners.
The training recipe combines high-resolution continual training, frame-based RL post-training, and one-step distribution-matching distillation, with bidirectional and streaming inference. It is designed to work with several base generators. In the project's benchmark, a five-second 1344×768 24 fps MiniMax H3 video dropped from 152.3 s at full resolution to 5.64 s with a two-stage pipeline (27.03× speedup) on NVIDIA GB200 GPUs.
Per a YouTube reviewer, output is sharper but can alter details, so it is not strictly faithful to the input; the same source says code has been released and borrows from the LTX Refiner. This was not confirmed on the fetched pages, so the repository, weights and license are not listed here.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Upscales low-res video to up to 4K in a single step and is model agnostic. Output is sharper but changes details, so it is not faithful. Code released, borrowed from LTX Refiner.
Closest neighbours by meaning, not just by category.
Model · NVIDIA
Open few-step diffusion refiner from NVIDIA that regenerates low-resolution H3 video drafts at 768P to 4K, 5–15 seconds.
Model · NVIDIA
NVIDIA's open-source image upscaler, v1.5, for latent, Flux and Z-Image outputs, targeting fast 2K results on consumer GPUs.
Model · Nvidia
NVIDIA's 5B and 14B hybrid linear-attention video diffusion transformers generating up to 720p on a single GPU.
Model
Streaming real-world video super-resolution using sparse large-model keyframes and a lightweight propagation network.
Model · Hao AI Lab
Open-weight 4-step sparse-distilled version of MiniMax H3 for fast text-to-video-and-audio generation on Blackwell GPUs.
Model · NVIDIA
NVIDIA research system where a code game engine drives a two-step streaming MiniMax-H3 video model for real-time rendering.
