Fast Preview & Optimize

Core ComfyUI's Preview Image writes a lossless PNG to the temp folder every single run. That's fine when you're just looking at the screen. It's genuinely annoying when the image is leaving your machine - over the ComfyUI API, or to a cloud workflow runner (the "Flow Worker"-style APIs this pack's README mentions) where the response payload and your bandwidth both scale with file size. That's the problem Fast Preview & Optimize exists to fix: same behavior as the stock preview node, but it re-encodes your image to WebP, JPEG, or PNG at a quality you choose, and then tells you the resulting file size right on the node. The name is a little grand, but it does what it says.

How it works

Under the hood it's refreshingly boring - which is a compliment. The node takes the image tensor, converts it to a PIL image, and saves it into ComfyUI's temp directory with a random 8-character filename prefix, exactly like the core preview node. The clever parts are the compression choices and the size readout.

WebP saves use method=2, which favors conversion speed over squeezing out the last byte - the README's "fast conversion" in practice. JPEG saves use optimize=True. PNG is where you should pay attention: the quality slider is remapped to zlib compress_level, so quality 100 means no compression - a giant file - and a low slider means max compression. "Quality" for PNG here is really "how hard to compress," not perceptual quality.

Then a small web extension listens for the node's file_sizes payload and injects a read-only widget showing something like Size: 186.42 KB. It only appears after the first run - there's nothing to measure before you've generated.

The inputs that matter

• images - the IMAGE tensor from your sampler, VAE decode, or upscale.
• image_format - webp (default), jpeg, or png. WebP is the sensible default for previews: good quality, small file. Pick JPEG if something downstream insists on JPEG.
• quality - integer 1–100, default 80. Fine-tune per format as above; for WebP, 80–90 is the sweet spot and 100 trends toward lossless-and-huge.

One more real-world gotcha: JPEG has no alpha channel, so the node converts RGBA images to RGB before saving - your transparency quietly disappears if you choose JPEG.

Installing it

The README says ComfyUI Manager support is "Coming Soon," which is polite for not yet - you won't find this in Manager's search today. The real install is the manual one:

cd ComfyUI/custom_nodes
git clone https://github.com/AabhasTech/ComfyUI_Fast_Preview.git

Restart ComfyUI and grab it from the image category. The pleasant surprise: there's no requirements.txt at all. It only needs PIL and numpy, which ComfyUI already ships, and there are no model files to download. This is a zero-dependency pack, which is more than most of the ecosystem can say - a nice change after the usual dependency-hell install dance.

Issues to expect

Nothing here is a mystery failure, but a few things will trip you up. The size widget is invisible until you've run the prompt once - don't go hunting for a config box. The temp folder is volatile, so those fast_preview_*.webp files get cleaned up on restart; that's the point (no disk bloat), but it also means don't use this node when you need a persistent artifact - that's what Save Image is for. And if you set PNG quality to 100 expecting the prettiest file, you'll instead get a near-uncompressed PNG that's probably bigger than the original. For quick previews and API payloads, leave it on WebP at 80 and let the size widget do its job.