Save Video

In ComfyUI there are two very different things you can want to save: an IMAGE batch (raw frames) or a VIDEO object (an actual container with a codec, maybe an audio track). The pack's Save Video by Image handles the first. Save Video handles the second - and its job is more interesting than it sounds, because a VIDEO object already has a format, and the worst thing a save node can do is re-encode it into something else.

This node saves a VIDEO input to your output directory preserving its source extension, optionally remuxes your prompt/workflow metadata into it, detects alpha and builds a playable preview, and returns the absolute path. No unnecessary re-encode of your video-gen output.

How it works

• video - the VIDEO object, optional. Leave it unconnected and the node just passes through empty.
• filename - default video/ComfyUI, a prefix that goes through ComfyUI's path generation (so %date:yyyy-MM-dd% placeholders work).
• auto_increment - numbered files (name_00001_.mp4) vs. overwrite.
• save_output - output directory when on, temp directory when off.
• save_metadata - remuxes the workflow/prompt JSON into the container's comment field using ffmpeg, with streams copied rather than re-encoded. Your video carries its own "workflow included" data, same as your PNGs.

The extension logic is the quietly impressive part: the node checks the source file's extension and keeps it - an MP4 saved by a loader stays MP4, a WebM stays WebM. If the source carries an alpha channel (detected via ffprobe), it also generates a WebM preview in the temp directory so you can actually see the transparency in the UI without opening a heavyweight file.

Output is file_path - the absolute path of the saved file.

Install

Part of ComfyUI-1hewNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/1hew/ComfyUI-1hewNodes

Restart ComfyUI. It shells out to ffmpeg/ffprobe, which ship with ComfyUI - no extra install.

Where it fits

End of any pipeline that ends in a VIDEO object: video generation models, Load Video + frame-processing + re-encode, URL-to-video downloads. If you want frames instead, that's Save Video by Image. The mental model is simple: IMAGE → Save Video by Image, VIDEO → Save Video. Mixing them up is the number one source of "why won't this connect" confusion.

Common issues

• The file "disappeared" - check save_output; when off it writes to temp, which is where previews live.
• It's re-encoding and you don't want it to - metadata remux copies streams; it shouldn't re-encode. If you're seeing a format change, the input itself was the changed format.
• Preview shows a flat image - alpha detection depends on the source probing correctly; if ffprobe can't tell there's alpha, you get the normal preview, not a transparency one. Cosmetic, not a data-loss issue.

For anyone who's lost a render to a "save" node that silently transcoded their carefully chosen format - this one respects the container, which is honestly the whole review.