The　degradation　types　of　snow　are　complex　and　diverse.　Existing　methods　employ　sophisticated　model　architectures　to　model　sufficient　visual　representations　for　snow　removal.　In　order　to　remove　snow　more　efficiently,　inspired　by　the　powerful　visual　representations　of　pre-trained　large　models　and　the　efficient　parameter　fine-tuning　paradigm　in　the　field　of　natural　language　processing,　we　have　pioneered　the　exploration　of　applying　efficient　parameter　fine-tuning　in　low-level　vision.　Taking　the　desnowing　task　as　the　starting　point,　we　introduced　TuneSnow,　a　framework　for　efficient　parameter　fine-tuning　that　can　be　integrated　with　desnowing　network　to　improve　desnowing　performance.　Initially,　we　introduced　Hybrid　Adapters　for　the　efficient　fine-tuning　of　pre-trained　vision　models.　We　then　proposed　a　Progressive　Multi-Scale　Perception　module　(PMSP)　to　harness　the　feature　representation　potential　of　pre-trained　models.　Finally,　we　presented　a　Degraded　Area　Restoration　module　(DAR)　based　on　Multi-Scale　Fusion　Refinement　module　(MSFR)　to　recovery　details　after　desnowing.　Extensive　experiments　demonstrate　that　our　approach　trains　only　15%　of　the　parameters　and　delivers　state-of-the-art　performance　on　multiple　publicly　available　datasets.　TuneSnow　can　serve　as　a　plug-and-play　component　to　enhance　the　performance　of　other　U-shaped　image　restoration　models,　including　derain,　dehaze,　deblur,　and　more.　The　code　and　datasets　in　this　study　are　available　at　https://github.com/dxw2000/PEFT-TuneSnow.　©　2024　Elsevier　Ltd