Save Video (FFMPEGA)

FFMPEGA's Agent renders to a temp file somewhere in ComfyUI's depths, and if you've ever gone hunting for that file, you know the pain this node kills. Save Video (FFMPEGA) takes a video path - typically the Agent's video_path output - copies the file into ComfyUI's output directory, and shows you an inline preview. Crucially, it does no re-encoding. It's a file copy with a preview, which means it's instant and lossless; the encode happened upstream in the Agent.

That no-re-encode detail is worth re-reading, because it's the difference between "add a Save Video node and lose a generation of quality" and "add it and lose nothing." It also writes a workflow PNG thumbnail alongside the video, so you can drag-and-drop the saved file back into ComfyUI later and reconstruct the workflow that made it.

The inputs that matter

• video_path - the file to save. Wire it from the FFMPEG Agent's video_path output (or Load Video Path). The file already contains its audio; it's copied as-is.
• filename_prefix - the saved filename's prefix, default FFMPEGA. Supports ComfyUI formatting like %date:yyyy-MM-dd%, so you can get date-stamped outputs for free.
• save_output - default On. When On it copies to the output directory; when Off it still shows the inline preview and emits frames/audio/path but doesn't write a copy. That's the "am I happy with this before committing" switch.
• overwrite - default Off, which means auto-incrementing counter when a file already exists. Flip it On to replace instead.

Outputs: images (the frames as a batched tensor), audio, video_path (the saved location - chain it into another loader or an editor), and mask_points pass-through.

Installing it

Part of ComfyUI-FFMPEGA: ComfyUI Manager → search ComfyUI-FFMPEGA → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AEmotionStudio/ComfyUI-FFMPEGA.git
cd ComfyUI-FFMPEGA
pip install -r requirements.txt

Restart ComfyUI. FFMPEG on PATH is the pack's one hard dependency (ffmpeg -version).

Where people get burned

The most common mistake is double-saving: leave the Agent's own save_output On and wire in a Save Video, and you get two copies and two previews. Turn the Agent's off - the README says as much - and let one node own the output. The second gotcha is the mental model: because it doesn't re-encode, this is not where you fix quality. If you want a smaller file or a different codec, do that as an effect in the Agent (compress/convert/quality skills), then Save Video just files the result. Think of it as the pack's mailbox, not its workshop.