Apps

Upscayl

Upscayl stores several Electron and GPU-related caches under its profile on Windows so the interface and AI upscaling workflow start faster after the first run. Those files include Chromium HTTP cache entries in Cache_Data, V8 compiled script data in Code Cache, GPUCache records, Dawn/WebGPU shader and pipeline cache data in DawnCache, plus accumulated application logs; Kudu removes those disposable caches and logs while leaving your saved images, model files, app settings, and project data alone.

Why it matters

Why clean Upscayl?

  • Invalidated V8 compiled code after an Upscayl or Electron update can make the app open slowly, show a blank window briefly, or feel stuck on first launch until the code cache rebuilds
  • Stale Chromium Cache_Data entries can keep old web assets around, leading to missing icons, partially rendered panels, or an interface that looks broken after an update
  • GPU driver updates often invalidate GPUCache data, and keeping the old cache can cause black previews, flickering UI elements, or a sluggish first render while graphics resources are recreated
  • Outdated DawnCache shader and pipeline blobs from a previous WebGPU or driver state can trigger stutter when opening the app, delayed preview rendering, or intermittent failures when the GPU path initializes
  • Large settled log folders grow quietly over time and waste disk space, especially if Upscayl has logged repeated startup or inference warnings that you no longer need
  • A corrupted cache file can send Upscayl into repeated slow starts or odd one-off UI glitches, and clearing the disposable caches forces a clean rebuild without touching your images or settings
What gets cleaned

Cache paths Kudu targets

Windows

%AppData%/Upscayl/Cache/Cache_Data
%AppData%/Upscayl/Code Cache
%AppData%/Upscayl/GPUCache
%AppData%/Upscayl/DawnCache
%AppData%/Upscayl/logs
Frequently asked

Common questions about Upscayl

Free & open source

Download Kudu and reclaim your disk space.

Available on Windows, macOS, and Linux. No account required, no feature gates, no telemetry without consent. All cleaning targets are open source and community-auditable.