Video Super-Resolution & Restoration¶
Video
Open Source¶
-
DUO-VSR: Dual-Stream Distillation for One-Step Video Super-Resolution Discovered by AI
2026
DUO-VSR is a dual-stream distillation framework that compresses multi-step video super-resolution models into a one-step generator. It utilizes progressive guided distillation initialization and joint optimization of distribution matching distillation and GAN losses to maintain high visual quality and temporal consistency.
distillationone-step-generationvideo-super-resolution -
LiteVSR: Lightweight Adaptation of Frozen Diffusion Transformers for Video Super-Resolution Discovered by AI
2026
LiteVSR is a lightweight video super-resolution framework that adapts frozen Diffusion Transformers using a state-aware adapter. It simplifies the adaptation process by predicting a constant velocity field across timesteps, reducing the need for complex time-varying transformations.
adapterdiffusion-transformervideo-super-resolution -
NanoVSR: Towards Real-Time Video Super-Resolution on Edge Devices Discovered by AI
2026
A scalable, fully convolutional video super-resolution architecture designed for resource-constrained edge devices. It utilizes structural reparameterization to collapse multi-branch blocks into standard convolutions during inference, achieving real-time performance without custom CUDA operations.
convolutionaledge-aireal-timesuper-resolution -
PS-SR: Pseudo-Single-Step Video Super-Resolution via Speculative Diffusion Discovered by AI
2026
A video super-resolution framework that accelerates diffusion-based VSR through speculative diffusion, using a powerful base model to guide a fast draft model for high-quality, single-step inference.
diffusionrestorationspeculativevsr -
SparkVSR: Interactive Video Super-Resolution via Sparse Keyframe Propagation Discovered by AI
2026
An interactive video super-resolution framework that allows users to guide and correct upscaling results using sparse keyframes as control signals. The system propagates user-refined keyframes across the video sequence to ensure temporal consistency and eliminate artifacts.
interactivekeyframesuper-resolutiontemporal-consistency -
SwiftVR: Real-Time One-Step Generative Video Restoration Discovered by AI
2026
A streaming, one-step generative video restoration framework designed for real-time live streams, reducing spatial attention complexity and autoencoder latency under a causal chunk-wise protocol.
one-stepreal-timerestorationvideo-sr -
DicFace: Dirichlet-Constrained Variational Codebook Learning for Temporally Coherent Video Face Restoration Discovered by AI
2025
A video face restoration framework that extends discrete codebook representations (VQ-VAE) into continuous Dirichlet-distributed variables. This formulation enables temporally coherent face restoration and detail recovery from highly degraded video inputs.
detail-recoveryface-restorationtemporal-coherencevq-vae -
DiffVSR: Revealing an Effective Recipe for Taming Robust Video Super-Resolution Against Complex Degradations Discovered by AI
2025
A robust video super-resolution framework that tames diffusion models to balance fidelity and temporal consistency, specifically optimized to handle complex and severe real-world degradations.
diffusionrestorationvideo-sr -
DLoRAL: One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution Discovered by AI
2025
A one-step diffusion model for real-world video super-resolution that utilizes a Dual LoRA Learning paradigm to enhance spatial details and maintain temporal consistency.
diffusionloraone-stepvideo-super-resolution -
DOVE Discovered by AI
2025
An efficient one-step diffusion model for real-world video super-resolution, designed to significantly reduce inference latency while maintaining high visual fidelity.
diffusionone-stepsuper-resolution -
FlashVSR Discovered by AI
2025
A real-time, diffusion-based one-step streaming framework for video super-resolution, utilizing locality-constrained sparse attention and a tiny conditional decoder to achieve high efficiency.
diffusionreal-timestreaming -
OASIS: Towards Redundancy Reduction in Diffusion Models for Efficient Video Super-Resolution Discovered by AI
2025
An efficient one-step diffusion model with attention specialization designed to reduce computational redundancy in video super-resolution. It leverages the content information preserved in low-quality inputs to streamline the denoising process, significantly lowering computational overhead while maintaining high visual quality.
diffusionefficiencyone-stepvideo-super-resolution -
RealisVSR: Detail-enhanced Diffusion for Real-World 4K Video Super-Resolution Discovered by AI
2025
A detail-enhanced video super-resolution framework that integrates a Consistency Preserved ControlNet with Wan2.1 and utilizes a High-Frequency Rectified Diffusion Loss to recover high-frequency details under complex real-world degradations.
4kcontrolnetdiffusionvideo-super-resolution -
SeedVR Discovered by AI
2025
A diffusion transformer framework designed for generic video restoration of arbitrary length and resolution, utilizing shifted window attention to maintain temporal consistency and high fidelity.
diffusion-transformervideo-restorationvideo-super-resolution -
SeedVR2 Discovered by AI
2025
A highly efficient one-step diffusion-based video restoration model trained via diffusion adversarial post-training, delivering high-fidelity upscaling for real-world videos.
diffusionone-steprestoration -
STAR: Spatial-Temporal Augmentation with Text-to-Video Models for Real-World Video Super-Resolution Discovered by AI
2025
A spatial-temporal augmentation framework that integrates pre-trained text-to-video (T2V) models into real-world video super-resolution, improving temporal consistency and detail generation.
diffusiont2vvideo-sr -
Stream-DiffVSR: Low-Latency Streamable Video Super-Resolution via Auto-Regressive Diffusion Discovered by AI
2025
A causally conditioned, low-latency video super-resolution framework that operates strictly on past frames using auto-regressive temporal guidance and distilled denoising.
diffusionlow-latencyvideo-sr -
UltraVSR: Achieving Ultra-Realistic Video Super-Resolution with Efficient One-Step Diffusion Space Discovered by AI
2025
An ultra-realistic video super-resolution model that leverages efficient one-step diffusion space to generate high-fidelity details while maintaining temporal coherence.
diffusionone-stepvideo-sr -
Vivid-VR: Distilling Concepts from Text-to-Video Diffusion Transformer for Photorealistic Video Restoration Discovered by AI
2025
A generative video restoration method built on a text-to-video diffusion transformer (DiT) and ControlNet. It addresses distribution drift and multimodal alignment issues by distilling concepts from the foundation model, achieving high texture realism and temporal consistency.
controlnetdiffusion-transformerphotorealisticrestoration -
Arbitrary-Scale Video Super-Resolution with Structural and Textural Priors Discovered by AI
2024
An arbitrary-scale video super-resolution framework that leverages structural and textural priors. It uses a flow-guided recurrent unit to aggregate spatiotemporal information and a scale-aware lookup table to handle arbitrary upscaling factors.
arbitrary-scalerecurrentrestorationvideo-super-resolution -
Beyond Alignment: Blind Video Face Restoration via Parsing-Guided Temporal-Coherent Transformer Discovered by AI
2024
A blind video face restoration model that utilizes a parsing-guided temporal-coherent transformer to handle complex real-world degradations. It integrates semantic parsing maps to guide the restoration of facial structures and maintain temporal consistency across diverse poses.
blind-restorationface-restorationparsing-guidedtransformer -
EvTexture Discovered by AI
2024
An event-driven framework for video super-resolution that leverages high-frequency details from event signals to enhance texture regions and recover fine details.
event-cameratexture-enhancementvideo-super-resolution -
Learning Truncated Causal History Model for Video Restoration Discovered by AI
2024
A video restoration framework that learns a truncated causal history model (TURTLE) to store and summarize historical frame representations, improving efficiency and performance in video deblurring and denoising.
causal-historydeblurringdenoisingvideo-restoration -
VEnhancer: Generative Space-Time Enhancement for Video Generation Discovered by AI
2024
A generative space-time enhancement framework that performs spatial super-resolution, temporal frame interpolation, and video refinement within a single unified video diffusion model.
diffusioninterpolationvideo-sr -
VideoGigaGAN: Towards Detail-rich Video Super-Resolution Discovered by AI
2024
A generative video super-resolution model that upscales videos with high-frequency details while maintaining temporal consistency by extending image-based generative upsamplers.
gantemporal-consistencyvideo-sr -
StableVSR: Enhancing Perceptual Quality in Video Super-Resolution through Temporally-Consistent Detail Synthesis using Diffusion Models Discovered by AI
2023
A video super-resolution method based on diffusion models that enhances perceptual quality by synthesizing realistic and temporally-consistent details. It introduces a Temporal Conditioning Module (TCM) to adapt pre-trained single-image super-resolution models for video.
diffusionrestorationtemporal-consistencyvideo-super-resolution -
Upscale-A-Video Discovered by AI
2023
A text-guided latent diffusion framework for real-world video upscaling and super-resolution, designed to maintain high visual fidelity and temporal consistency.
diffusiontemporal-consistencyupscaling -
RVRT Discovered by AI
2022
A recurrent video restoration transformer that balances parallel and recurrent architectures, utilizing guided deformable attention for efficient clip-to-clip alignment and temporal fusion.
recurrentrestorationtransformer -
VRT Discovered by AI
2022
A parallel video restoration transformer designed for video super-resolution, deblurring, and denoising, featuring multi-scale temporal mutual self-attention to capture long-range dependencies.
denoiserestorationtransformer -
BasicVSR++ Discovered by AI
2021
An advanced recurrent video super-resolution framework that introduces second-order grid propagation and flow-guided deformable alignment to effectively exploit spatiotemporal information across frames.
alignmentrestorationsuper-resolution -
RealBasicVSR Discovered by AI
2021
A real-world video super-resolution framework that balances the tradeoff between detail synthesis and artifact suppression using an image pre-cleaning stage and a recurrent propagation structure.
real-world-vsrvideo-restorationvideo-super-resolution -
BasicSR Discovered by AI
2020
An open-source image and video restoration toolbox based on PyTorch. It supports various state-of-the-art architectures for super-resolution, denoising, and deblurring, including EDVR, BasicVSR, and SwinIR.
super-resolutiontoolboxvideo-restoration -
Waifu2x-Extension-GUI Discovered by AI
2020
A powerful, user-friendly GUI application for upscaling and interpolating videos, images, and GIFs, integrating multiple state-of-the-art algorithms such as Waifu2x, Real-ESRGAN, Real-CUGAN, and RTX Video Super Resolution.
guirestorationupscalingwaifu2x -
Video2X Discovered by AI
2018
A popular open-source video super-resolution and frame interpolation framework that integrates state-of-the-art algorithms like Waifu2X, Real-ESRGAN, Real-CUGAN, and RIFE.
animeinterpolationupscaling
Proprietary / Industry Tools¶
-
AVCLabs Video Enhancer AI Discovered by AI
2024
An AI-powered video enhancement and upscaling software that restores clarity to blurry, noisy, or low-resolution videos. It utilizes deep learning models to upscale footage up to 8K, remove compression artifacts, and enhance facial details.
commercialdenoisingrestorationupscaling -
Topaz Video AI Discovered by AI
2024
A professional-grade desktop software utilizing temporally aware AI models to perform video upscaling, enhancement, stabilization, and motion smoothing.
proprietary-toolvideo-enhancementvideo-upscaling -
NVIDIA RTX Video SDK Discovered by AI
2023
An AI-powered video upscaling and artifact reduction SDK that leverages NVIDIA RTX Tensor Cores to sharpen edges, restore features, and remove compression artifacts in real time.
real-timertxsdkupscaling -
Pixop Discovered by AI
2020
A cloud-based, ML-powered video enhancement and restoration platform designed for broadcasters and media companies, offering automated upscaling, denoising, and SDR-to-HDR conversion.
cloudrestorationupscaling
