🎯 Fast Unsharp Sharpen

Most "AI video is soft" complaints are really "my encoder crushed the detail" complaints. And most sharpening advice - slap an upscaler on it - is overkill when all you want is a bit of edge definition back. FastUnsharpSharpen is the boring, correct tool: unsharp masking, the same trick Photoshop has used for decades, wrapped in a node that runs over a whole frame batch without melting your GPU.

Unsharp masking works by blurring a copy of the image, subtracting it from the original to find the detail layer (the high-frequency edges), then adding a scaled version of that detail back onto the original. The result is edge contrast without the halo-filled mess you get from naive convolution sharpening. The pack calls it "simple and efficient" and that's the accurate review - it's not a magic enhancer, it's a clean detail restorer.

The inputs that matter

• images - your frame tensor (IMAGE).
• strength - 0.0 to 2.0, default 0.5. This is how much of the detail layer gets added back. 0.5 is a sensible "I noticed it was soft" amount; anything above 1.0 starts getting crunchy on video.
• use_gpu - off by default, and that's the right call. The CPU path is fast enough for frames at this resolution, and keeping it off CPU means you don't spike VRAM mid-pipeline. Flip it on only if CPU becomes the bottleneck and you have GPU headroom.

Output is a single IMAGE tensor of identical shape, so it slots into any enhancement chain. There's nothing else to configure - no radius, no threshold. That's a feature. For video you generally don't want to micromanage sharpen parameters per-frame; you want a consistent pass.

Install

It's part of the VRGameDevGirl pack:

cd ComfyUI/custom_nodes
git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl

Or search vrgamedev in ComfyUI Manager, install, restart. Zero extra dependencies beyond what ComfyUI ships.

Where it fits

Use it after decode and before encode in your video pipeline, or as a final pass after upscaling to claw back edge definition the upscaler's detail pass smeared. It pairs well with the pack's other enhancement nodes - grain first, then sharpen, so you're sharpening actual detail rather than noise. (Sharpen before grain and you amplify noise, which is a classic beginner trap.)

If you need something punchier, the pack also has FastLaplacianSharpen (aggressive edge boost) and FastSobelSharpen (directional gradient emphasis). But unsharp is the one I'd reach for first on general footage: it's the least likely to leave ugly artifacts, and at default strength it's basically risk-free. The trap is cranking strength past ~1.0 expecting more detail and getting halos instead - video sharpening has diminishing returns fast, and this node correctly starts you low.