Video Save

If you've built a video graph with ComfyUI-tbox and made it to the end, this is the node that actually writes the file. Feed it frames, get an mp4 or webm on disk - same job as Video Combine in other packs, just tbox's own take on it, with its own format list and (if you wire one up) a way to encode long clips in chunks instead of blowing your VRAM budget on the last step.

How it works

VideoSaver takes a stream of image frames - plus an audio track, if you have one - and encodes them to the path you give it. You pick the container/codec, a quality level, whether the clip should ping-pong (play forward then reverse, a cheap trick for a loop with no visible seam), and the frame rate. It's a pure output node: nothing downstream reads from it, it's where the graph ends.

The inputs and outputs that matter

• path (required STRING) - where the file gets written.
• format (required enum) - five options: video/h264-mp4, video/h265-mp4, video/nvenc_h264-mp4, video/nvenc_h265-mp4, video/webm. The nvenc_ variants push the encode onto your GPU's hardware encoder, which is much faster than software encoding - but it's a real trade, not a free lunch. People comparing the two on identical ComfyUI video renders consistently report the software (CPU) h264/h265 path looking noticeably cleaner on detailed footage, and specifically recommend switching off nvenc for anything you're actually going to publish, saving it for quick previews only.
• quality (required enum, default 100) - a 50–100 scale; higher keeps more detail at a bigger file size.
• pingpong (required BOOLEAN, default false) - plays forward then backward for a seamless loop.
• images (optional IMAGE) and audio (optional AUDIO) - your rendered frames and, if you've got one, a synced audio track to mux in. Forget to wire audio and you'll get a silent video with no error.
• frame_rate (optional INT/FLOAT, default 25, 1–60) - playback fps.
• meta_batch (optional, type BatchManager) - plug tbox's own BatchManager node in here and VideoSaver processes your frames in chunks instead of holding the whole clip in VRAM at once. If you're saving anything long, this is the pairing to reach for.

No outputs - like any save node, it's a dead end on the graph, which is fine, that's the job.

How to install it

Through ComfyUI Manager: search ComfyUI-tbox, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ai-shizuka/ComfyUI-tbox

then restart. Worth knowing going in: the pack's own README is essentially empty - three bare section headers (Image Node / Video Node / ControlNet PreProcessor) and nothing written under any of them, so there's no documented dependency list to check against. In practice, video encoding out of ComfyUI leans on ffmpeg being present and on PATH; if saves fail silently or error right after ComfyUI starts, that's the first thing worth checking on a fresh machine.

Common issues & troubleshooting

nvenc output looks soft or blocky next to what you previewed in the UI. Not a bug, just the hardware-encoder trade-off. Switch to h264-mp4 or h265-mp4 for your final export and keep nvenc for fast iteration.

Long renders OOM partway through encoding. Wire a BatchManager into meta_batch and set frames_per_batch to something your card can actually hold (16 is the default, and a reasonable starting point). Encoding in chunks instead of all at once is the entire reason that node exists.

No audio in the output even though you thought you added it. audio is optional and easy to leave dangling - VideoSaver won't pull a track from anywhere else in the graph, it has to be wired explicitly.

webm won't open in your editor. VP8/VP9-in-webm plays fine in a browser but isn't universally supported by desktop editors. If you need broad compatibility (Premiere, DaVinci, etc.), pick one of the mp4 options instead.