Video Super-Resolution (VSR) has benefited significantly from large-scale pre-trained video generators, which provide powerful priors for realistic detail synthesis. Existing methods, however, rely on fine-tuning the generative backbone, incurring substantial computational costs and risking catastrophic forgetting of learned priors. We reconsider how these priors can be exploited from a frequency perspective: pre-trained generators are inherently capable of synthesizing high-frequency details given structural guidance, while low-quality videos supply reliable low-frequency information. Since low-frequency content is largely domain-agnostic, a frozen generator can perform VSR directly when the input structure is properly aligned to its embedding space. Building on this insight, we propose LiteVSR, a minimalist framework that performs VSR using a completely frozen Diffusion Transformer (DiT) with a lightweight State-Aware Adapter. The adapter employs a dual-stream architecture that jointly processes static structural cues from the low-quality input and dynamic cues from intermediate denoising states through time-dependent cross-attention, enabling adaptive transition from structural alignment to texture refinement as denoising proceeds. LiteVSR achieves state-of-the-art restoration quality with only 12.68% trainable parameters and 12 GPU-hours of training on a single A100, while preserving compatibility with off-the-shelf fast sampling algorithms.
