😼> Vignette

Real camera lenses darken toward the edges of the frame - it's just optics, light falls off faster at the corners than the center. Diffusion models almost never generate that on their own, which is one of the small tells that an image is AI-made even when everything else about it looks convincing. Vignette is the node that fakes it back in: it darkens (or lightens) the corners of your image after generation, purely as a post-processing pass over pixels that already exist.

It ships as part of a small post-processing suite the YANC pack added in August 2024 - sitting alongside Bloom, Film Grain, HUE, Blur, Sharpen and a few others, all doing one specific cosmetic job each. Vignette's job is corner falloff, full stop. It doesn't touch generation, doesn't need a mask, doesn't need anything upstream except an image.

What it actually takes

Only four inputs, and three of them you'll actually touch:

• image - whatever came out of your VAE Decode or an earlier post-processing node in the chain.
• intensity (default 0.25, range -1 to 1) - how strong and how far-reaching the falloff is. Push it up for a heavier, more noticeable vignette; back it toward 0 for something you'd only notice if you went looking.
• opacity (default 1, range 0–1) - a straight blend control between the original image and the vignetted version. Drop it if intensity gives you the shape you want but it's too strong overall.
• color - black or white. Black is the classic photographic vignette - moody, cinematic, the look you're used to from real lenses. White pushes the corners toward blown-out/faded instead, more of a hazy, overexposed-film feel. There's no in-between option; pick the one that matches what you're going for.

Output is a single image, ready to feed straight into Save Image or into the next node in your post-processing stack.

Where it fits in a real workflow

Don't treat this as a magic realism button on its own - it's one ingredient. If you're chasing "this doesn't look AI-generated," the community consensus on post-processing checklists (grain, mild desaturation, a touch of color cast, lens-style artifacts) treats vignetting as one line item among several, not the whole fix. Stack it with Film Grain and a small saturation pull from this same pack - or from whatever post-processing nodes you already have - rather than expecting the vignette alone to sell the shot. It's also cheap enough that there's no real cost to leaving it in a workflow at a low intensity as a default finishing touch.

Installing it

Same pack, same install path as every other YANC node: search ComfyUI_yanc in ComfyUI Manager and install, or clone it by hand - cd ComfyUI/custom_nodes && git clone https://github.com/ALatentPlace/ComfyUI_yanc - then restart ComfyUI. No models to download, no extra dependencies called out in the README; it's a plain image-processing node, so once the pack is installed and ComfyUI restarts cleanly, it just works.

If something looks off

There isn't much documented that goes wrong with this one specifically - it's a simple, deterministic image filter, not something with network calls or model weights that can fail. The two things actually worth checking if the output doesn't look like you expect: make sure opacity isn't accidentally left at 0 (the node will run and pass the image through with no visible effect), and remember intensity can go negative - if your corners look wrong rather than just "off," check you haven't dragged it below 0 by mistake while dialing it in. Beyond that, if the node fails to appear at all after install, it's the same story as the rest of the pack: check the ComfyUI console on startup for an import error, and confirm the repo landed directly inside custom_nodes/ rather than nested a folder too deep.