Vivid Sharpen

Renders and upscales come out soft. That's normal, and the fix is not another diffusion pass - sharpening is a blur and a subtraction, it costs milliseconds, and it doesn't rewrite anyone's face. Reach for it before you reach for img2img.

Vivid Sharpen is the pack's take on that. The mechanism: blur a copy of the picture, invert it, and blend that high-pass layer back over the original in vivid-light mode. Edge pixels gain local contrast, flat pixels cancel out and stay put. What you see reads as detail rather than as the bright halo an ordinary unsharp mask leaves along every strong boundary. The module even keeps light above 1.0 alive through the stack and carries a fourth channel untouched, so an alpha doesn't get chewed.

Three widgets. That's the whole node.

The inputs

• images - takes a batch, and processes it a frame at a time, so a whole video sequence can go through one node without you babysitting memory.
• radius (default 1.5) - the size of the detail being picked out. Around 1.0 accents fine texture, skin, fabric weave. At 5.0 and up you're accenting broad shapes, which at some point stops being sharpening and becomes contrast. This is the control people get wrong by leaving it high.
• strength (0.0–1.0) - how much of the sharpened version is mixed back over the original. 0.0 is a no-op, 1.0 is the sharpened result outright.

Single output: images, same size, same batch order.

The honest advice on settings

strength defaults to 1.0 and that's the hero setting, not the useful one. On a real render, 0.3 to 0.6 is where this node earns its keep - a pass nobody notices as a pass, which is the point of sharpening. Crank it and you'll get the crunchy, over-processed look that makes people decide they don't like sharpening nodes.

Same story with radius. Start at 1.0–1.5, look at 100%, and only raise it if texture is genuinely absent rather than just quiet.

The other thing worth saying: sharpening is not detail recovery. If the sampler didn't put fabric weave in the file, this node isn't going to find it. It accents what's there. When you want new detail, that's a generative upscaler's job and it comes with different risks - see the upscaling notes on the restoration-versus-generation split.

Install

ComfyUI Manager → search WAS Node Suite v3 → install → restart. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/was-node-suite-comfyui.git

Nothing gets installed, built or downloaded by the pack - ComfyUI 0.14.0+ and Python 3.10+. First start writes config.yaml and the pack's own folders under your ComfyUI user dir and takes a second or two longer than later starts, because it compiles to bytecode. Vivid Sharpen is part of the extras feature group, which is enabled by default.

Where people get burned

It won't import, or the node is missing. In v2 that was the pack's signature failure - ComfyUI would update a shared package and the pack's pinned version fought it, which is what all those 2024 "Import Failed" threads were about. v3 ships no dependencies, so there's nothing left to conflict. If the node specifically isn't in the menu, look at features.extras in config.yaml - set it to false and you get the standalone WAS Extras pack's nodes instead of these.

Haloes along hard edges. Vivid light is gentler than a plain unsharp mask, not immune. Bring strength down first, radius second. If you've already tried both and still see them, Vivid Sharpen (V2) has a second blur specifically for suppressing that halo.

The image comes out too contrasty. That's radius too high, not strength. Large radii accent broad tonal structure, which is contrast wearing a hat.

Nothing happened. strength is 0, or something upstream already delivered a float image outside the 0–1 range and the accents cancelled. Restore the default and re-run before digging further.