Video Combine Plus 🎥🔊⏯️

Every video workflow in ComfyUI ends the same way: a batch of frames leaves the sampler, and you need to glue them into a file. That's the job of the Video Combine family of nodes, and Video Combine Plus 🎥🔊⏯️ is a single-node fork of the classic one from VideoHelperSuite with a few genuinely useful extras bolted on - audio volume control, a one-click boomerang loop, and a companion PNG that drags your whole workflow back into the editor.

Positioning first: this isn't a big ecosystem pack. It's one node from one author (peterducan-hub), it has essentially zero community footprint, and if you already run VideoHelperSuite you can skip it - the base functionality is the same job. What it does better than the stock VHS "Video Combine" is bundle the fiddly stuff you'd otherwise wire up by hand.

How it works

You feed it an IMAGE batch (the frames your sampler produced - the same thing Wan or AnimateDiff spits out before a save node) and it pipes them straight into ffmpeg as raw RGB frames, then encodes the file on disk. No Python video library involved - the one real dependency is the ffmpeg binary itself, which is why the pack ships zero requirements.txt entries. It probes your ffmpeg build and tries hardware encoders in order - NVENC on NVIDIA, AMF on AMD, QSV on Intel, VideoToolbox on Mac - falling back to CPU libx264 if none work. It caches the first encoder that succeeds, so only your first run is noisy with probing. mp4 gets +faststart so it streams in a browser instead of loading fully first.

Audio is the rarest feature here. Feed it an AUDIO input (the type ComfyUI's LoadAudio produces) and it writes a temp WAV, then muxes it in with ffmpeg - AAC for mp4/mov, Opus for webm, MP3 for avi. No separate "combine video + audio" node required.

The inputs that matter

You'll touch four things regularly:

• images - your frame batch. The only required socket.
• frame_rate - 24 by default, and honestly this is where people trip: set it to match your model's native fps, not what looks nice.
• format - mp4/webm/mov/avi/gif. gif is a special case: it gets downscaled to 480px wide at 15fps with a palette pass, and audio is silently dropped.
• volume - the audio multiplier (0 mutes, 1 original, 2 doubles). This is the differentiator; VHS doesn't have it built in.

The rest are set-and-forget. pingpong appends your frames in reverse for a seamless boomerang loop (a real nicety for avatars and loading animations). save_metadata (on by default) writes a PNG of your first frame next to the video with the full workflow embedded as metadata - drag it back into ComfyUI to restore the graph. It's the same trick the built-in SaveImage node uses, and it's a small lifesaver when you find a three-month-old video and want to know what made it.

Output

It's an output node, so it finalizes the graph. The single output, Filenames, is typed VHS_FILENAMES - compatible with VideoHelperSuite's convention, so it drops into workflows that expect that type. Files land in your output folder with standard counter naming: VideoCombinePlus_00001.mp4.

The other "output" is cosmetic but notable: the custom JS preview replaces the boring preview with a full player - seek bar, loop, volume, download, and a "save current frame as PNG" button. That's the "video timeline" and "save current frame" the README brags about, and it's great for eyeballing a render without leaving ComfyUI. Note the inline preview only works for mp4, webm, and mov - avi and gif won't play in the node.

Install

The README's entire install story is two steps:

cd ComfyUI/custom_nodes
git clone https://github.com/peterducan-hub/Comfyui_VideoCombine_Plus

Then restart ComfyUI. That's it - no models, no Python deps. ComfyUI Manager may or may not index this pack (it's a small unregistered repo, so don't count on search); the clone is the reliable route. The one real requirement is ffmpeg on your PATH - the node hard-fails with "ffmpeg not found" if it isn't, and on Windows that's a PATH problem, not a ComfyUI one.

Where people get burned

The frame-rate mismatch above is the classic one - stuttery or too-fast motion almost always means your frame_rate doesn't match the model's native fps. Second, if you see console logs like "Trying encoder: h264_nvenc…" on first run, that's not an error - it's the encoder probe working, and it'll settle on whatever your hardware supports.

It's a small, opinionated utility, but it's genuinely useful for the one thing it does. If you want volume control, an instant loop, and a portable workflow baked into every render without adding VHS's whole surface area, this is a tidy way to get it.