How to blur images at 20,000 FPS (or 50 microseconds)
Acerola demonstrates a blur pipeline that can run at roughly 20,000 FPS, framing blur as a useful shader exercise rather than a throwaway post-process. The video walks through the core building blocks: box blur, race conditions, Gaussian blur, separable convolutions, GPU texture sampling, Kawase blur, upsample blur, and dual Kawase blur.
For engine and rendering developers, the practical takeaway is that blur quality and cost are tightly linked to how many samples you take, how you structure passes, and whether you can exploit separability or pyramid-style approaches. That makes the topic relevant well beyond a toy effect, especially for UI, bloom, depth-of-field, and other real-time filters where latency and bandwidth matter.
The video also points viewers to a paper on dithering.rocks and includes the accompanying Godot code on GitHub, which makes it easier to inspect the implementation details or adapt the approach to another renderer. Acerola notes that no generative AI or LLMs were used in making the video.
“How to blur images at 20,000 FPS (or 50 microseconds)”
- what
- Acerola shows how to blur images at about 20,000 FPS, or 50 microseconds per frame.
- who
- Acerola, with code shared from the AcerolaFX-Godot project.
- when
- Published as a new video; no specific date was provided in the source material.
- impact
- Useful for graphics programmers working on post-processing, UI effects, bloom, and other real-time filters.
Useful technical breakdown with practical shader takeaways
Follow shaders updates
See relevant stories in your personalized news feed.
Discussion