taco-group/SparkVSR

[ECCV 2026] SparkVSR: Interactive Video Super-Resolution via Sparse Keyframe Propagation

What it solves

SparkVSR addresses the "black box" nature of traditional Video Super-Resolution (VSR), where users have no way to correct artifacts or guide the restoration process. It transforms VSR into an interactive process by allowing users to provide a small set of high-quality keyframes as a control signal to guide the restoration of the entire video sequence.

How it works

SparkVSR uses a two-stage training pipeline built upon the CogVideoX1.5-5B-I2V base model:

  1. Latent-Space Adaptation (Stage 1): The model learns to fuse low-resolution (LR) video latents with sparsely encoded high-resolution (HR) keyframe latents to enable robust propagation of details across the video.
  2. Pixel-Space Refinement (Stage 2): The model refines perceptual details in pixel space, balancing adherence to the provided keyframes with the ability to perform "blind" restoration when keyframes are missing or imperfect.

At inference, users can provide keyframes via commercial APIs, open-source models (like PiSA-SR), or use a reference-free mode. The system propagates these high-quality priors across the video while remaining grounded by the original LR video's motion.

Who it’s for

This tool is designed for researchers and developers working on video restoration, as well as creators who need controllable, high-quality upscaling for tasks like old-film restoration and video style transfer.

Highlights

  • Interactive Control: Allows manual specification of keyframes to guide the super-resolution process.
  • Flexible Reference Modes: Supports API-driven, model-driven (PiSA-SR), or reference-free (blind SR) restoration.
  • Temporal Consistency: Improves temporal stability and restoration quality over existing baselines.
  • Generic Framework: Can be applied to tasks beyond simple upscaling, such as style transfer and old-film restoration.
  • ComfyUI Integration: Includes a ComfyUI implementation for more flexible, node-based inference.

Related

  • Project
    OpenImagingLab/FlashVSR

    FlashVSR is a one-step streaming diffusion framework for real-time video super-resolution that achieves high-speed upscaling and scalability to ultra-high resolutions.

  • Project
    ModelTC/LightX2V

    LightX2V is an open‑source inference framework for fast image and video generation (T2V, I2V, T2I, I2I, T2AV, etc.). It supports many state‑of‑the‑art models (MiniMax‑H3, Wan, HunyuanVideo, Qwen‑Image, etc.) and provides speed‑up techniques such as 4‑step distilled LoRAs, FP8/NVFP4 quantization, tensor/sequence parallelism, and disaggregated deployment on a wide range of hardware. Benchmarks show up to 3.9× faster inference than competing frameworks on H100 GPUs. The project offers Docker images, pip install, Gradio/ComfyUI front‑ends, extensive docs, and an online demo.

  • Project
    nv-tlabs/NKSR

    Neural Kernel Surface Reconstruction (NKSR) is a method for creating high-quality 3D implicit surfaces from large-scale, sparse, and noisy point clouds, capable of handling millions of points in seconds.

  • Project
    amanchadha/iSeeBetter

    A spatio-temporal video super-resolution project that uses Recurrent-Generative Back-Projection Networks and GANs to upscale low-resolution videos while maintaining temporal consistency and fine detail.

  • Project
    hao-ai-lab/FastVideo

    A unified post-training and real-time inference framework designed to accelerate video generation through sparse distillation and attention optimizations.