RAM Preview Image

Every ComfyUI graph eventually becomes a chain of hidden middle steps, and the way you inspect them is usually by wiring in a Save Image node - which dumps PNGs to disk for every batch, every run, until your output folder is a landfill. RAM Preview Image is the node that just looks. It shows the image in the node's preview panel, writes nothing to disk, and moves on.

The name is literal: the image is encoded to a base64 PNG in RAM and pushed to the UI over the websocket. Nothing touches your drive. For an intermediate that you're only checking once, that's exactly the behavior you want - you get visibility without the cleanup chore.

The one input, and why that's enough

• images - an IMAGE tensor, batch included. Feed it a whole batch and the preview cycles through the frames. It auto-downscales anything over a megapixel before encoding, so even a 2K intermediate previews without choking the browser.

That's it. It's an output node (no outputs, nothing to wire onward), which is correct - it's a terminal inspection point, like a scope probe on a circuit.

When you'd reach for it

• Debugging a pipeline. Put one after a mask-to-image conversion, a color-grade, or a crop you're unsure about, check it once, leave it in the graph. Zero ongoing cost.
• Batch inspection. The WtlNodes pack's apply_type live-preview system uses the same RAM-preview plumbing, so if you've used those color nodes you already know the pattern.
• Saving your output folder from itself. Keep Save Image only at the real endpoints; use RAM Preview for everything you're just eyeballing.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Scorpiosis0/ComfyUI-WtlNodes.git

Restart ComfyUI, or grab the pack via ComfyUI Manager (search "WtlNodes"). Nothing else to install.

Gotchas

• "RAM" here means the browser gets the image; it's not free. Large batches still serialize and push a base64 blob per frame. A 64-frame batch of 4K frames will make the UI sluggish even though the disk stays clean. Preview modest batches, or keep the inspection node behind a batch that isn't huge.
• It's a preview, not a save. If you wire it where you meant Save Image, you get nothing on disk. Which is the entire point, but it's a common confusion when copying someone else's graph.
• The preview isn't a workflow output. It doesn't show up in "completed" history the way a Save does, so don't rely on it for archival or for sharing results.