Save Image (WebP)

You've got a batch of 500 images and your output folder is eating your drive alive. The stock SaveImage dumps PNGs, and PNG is brutal on disk. Save Image (WebP) from the ComfyUI-bevvy pack is a drop-in replacement that writes WebP instead - typically a third to half the file size at the same visual quality - with quality, lossless, and compression-speed controls the built-in node never gives you.

It's an output node, the end of the line. Wire your IMAGE tensor into images, hit Queue, and files land in ComfyUI/output just like stock. No models to download, no heavy dependencies - under the hood it's a thin wrapper around Pillow's WebP encoder, which is why the whole thing is one small Python file.

What you actually set

The node has seven inputs, but three are what matter:

• quality (default 90, range 1–100) - the lossy quality knob. For sharing on the web, 80–90 looks identical to the original and is the sweet spot. You rarely need 100; you'll never see the difference.
• lossless (default off) - flips the encoder into lossless mode. Real gotcha: in lossless mode Pillow ignores quality entirely, so if you flip this on and crank quality to 100, you've changed nothing. You're trading a bigger file for a pixel-perfect one - only worth it for things like line art or sprites you might re-edit.
• compression (slow / normal / fast, default normal) - this maps to Pillow's WebP method, a 0–6 effort slider. Slow (method 6) squeezes the file harder at the cost of encode time; fast (2) is quick and chunky. "Normal" is the one you'll leave it on. If a batch is taking forever, dropping to fast is the cheap win.

The rest are fire-and-forget. filename_prefix behaves like SaveImage's - your files come out as prefix_00001.webp with the usual counter. embed_workflow (default on) stuffs the prompt and the full workflow graph into the file's EXIF metadata, so the "workflow included" culture ComfyUI built around PNG metadata mostly survives. Note the caveat: the drag-an-image-back-into-ComfyUI trick is bulletproof with PNGs and a little more format-dependent here, so for the ironclad reproducible share, keep a PNG around too.

The one input nobody else has is save (default on). Flip it off and the node writes to ComfyUI's temp directory under a random prefix instead of your output folder. That's a preview switch - render to the UI without cluttering your output, then turn it back on when you actually want files.

Installing it

Through ComfyUI Manager, search "ComfyUI-bevvy" (or "Save Image WebP") and hit Install, or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/calibancode/ComfyUI-bevvy

Then restart ComfyUI. The requirements.txt lists only numpy and Pillow - both ship with essentially every ComfyUI install, so there's nothing to fetch and no model files anywhere. If the node fails to load with an import error, that's the one thing to check:

cd ComfyUI/custom_nodes/ComfyUI-bevvy
pip install -r requirements.txt

When to use it (and when not to)

Reach for this when disk space or upload size matters: big batches, tiles for upscaling, anything you're shipping to a website or a Discord folder. The encoder is fast at "normal," the quality-for-size ratio beats PNG hands down, and it doesn't fight your workflow - it's the same output-folder rhythm as SaveImage.

Skip it when you need pixel-perfect output with workflow round-tripping as the priority. PNG stays the format of record in this ecosystem for a reason; treat WebP as the shareable, compact version of the same image, not a replacement for the archive. One pack, one node, does exactly one thing - and for a surprisingly large chunk of your saving, it's the right thing.

ComfyUI-bevvy is a small MIT-licensed pack by Caliborn; this is its only node. There's an argument the stock ComfyUI should just ship a WebP saver, but until it does, this is the 30-second answer.