Show HN: FlashVSR – AI Video Upscaler for AI-Generated and Low-Res Videos

AI-generated video has made remarkable progress, but the output often caps at 480p or 720p—blurry textures, jagged edges, and lacking detail. For creators, that’s far from production-ready. FlashVSR is a super-resolution tool designed specifically for this use case. Built on diffusion models, it delivers high-quality, high-efficiency 4× upscaling, ideal for enhancing generated video, restoring legacy footage, or any workflow demanding high resolution.

Key Technical Highlights

Single-Step Diffusion: Distills multi-step diffusion into a single inference step, significantly improving speed

Local Sparse Attention: Efficiently handles high-resolution inputs while avoiding texture repetition and positional drift

Lightweight Conditional Decoder: Reconstructs detail by conditioning on the original low-res frame, improving temporal stability and reducing flicker

Native Streaming Video Support: Designed for video-first input, balancing speed and output fidelity at 2K / 4K resolution

Who It’s For

Content creators working with generative outputs (e.g. Runway, Sora)

Developers of video enhancement tools

Restoration workflows: archival footage, film cleanup, AI reprocessing pipelines

Video AI is moving fast, but resolution shouldn’t lag behind. FlashVSR currently offers one of the best trade-offs between speed and quality among diffusion-based video upscaling models.