Generative video super-resolution runs fast on consumer GPUs
High-Throughput Generative Video Super-Resolution on a Single Consumer GPU
Computer Vision and Pattern Recognition
Summary
Improving video resolution usually requires a lot of computing power, making it hard to use on common computers. The authors developed a new method that boosts video quality using existing tools without retraining, running smoothly on a single high-end consumer GPU. Their approach upgrades the video frames in two steps: sharpening each frame first, then filling in smooth transitions between them. This way, they achieve better quality faster, though with some limits in handling complex motion.
What this means in practice
- •For video streaming services: Deliver higher-definition video streams by enhancing video quality efficiently on user devices with a single consumer GPU.$Commercial implications: Enables consumer-level devices to upscale video streams in real time, improving user experience without costly hardware upgrades.
- •For visual effects teams: Enhance the quality of HD video footage by applying fast super-resolution and smooth frame interpolation without retraining models.
Authors
Georgios Panaretou, Sotirios Chatzis
Abstract
Generative video super-resolution (VSR) can recover visually compelling high-frequency detail by leveraging diffusion priors, but iterative sampling and heavy spatio-temporal coupling make practical deployments difficult. We study a different operating point: practical HD generative VSR on a single high-end consumer GPU without end-to-end retraining. To this end, we present a zero-shot framework built entirely from pretrained components. The proposed system decouples spatial enhancement from temporal stabilization. A time-distilled one-step diffusion backbone handles spatial upscaling, while temporal coherence is promoted by a lightweight pretrained flow-based interpolator that synthesizes intermediate frames between super-resolved keyframes. To further improve scalability, we introduce a resolution-aware hybrid attention scheme that, in a zero-shot manner, replaces selected high-resolution global-attention layers with neighborhood attention while preserving global attention in semantically critical low-resolution layers, while also improving temporal consistency. Experiments show that the proposed framework reaches up to ~32 FPS at 720p resolution under interleaved inference on a single NVIDIA RTX 4090. At 4K, the proposed attention substitution can yield up to a 44% speedup over the original attention implementation. Across standard VSR benchmarks, the method consistently shifts the speed-quality trade-off decisively toward runtime efficiency, trading reconstruction fidelity for strong no-reference perceptual quality, while remaining reasonably competitive against generative approaches in temporal consistency. It also exposes clear limitations in handling more challenging motion, which a simple motion gate partially mitigates. Code will be released upon acceptance.