VideoBasic Merge Video

Concatenating videos in ComfyUI is usually a mess. Either you're stuck doing it in a file manager after a run, or you're stringing frame-tensor nodes together and re-encoding everything through the GPU for no reason. VideoBasicMergeVideo skips both: you hand it file paths, it hands the job to FFmpeg, and you get one merged mp4 in your output folder. No frames ever enter VRAM, which keeps it consistent with the rest of this pack.

It's from ComfyUI-VideoBasic, the same pack as the streaming upscale node - a small set of disk-based video utilities built around the idea that you shouldn't load whole videos into memory. Merge is the utility in the set: up to ten clips in, one file out.

How it works

The node builds an FFmpeg concat filter across all your inputs and re-encodes to H.264 (libx264, CRF 23) with AAC audio at 128k. Re-encoding rather than stream-copying is deliberate - it's what lets inputs of different formats, resolutions and codecs be glued together in one pass. If the concat filter fails for any reason, it falls back to the FFmpeg concat demuxer with a file list and -c copy (the pure stream-copy path, which only works when all inputs share codec and dimensions).

use_gpu is a lie the node hedges against: it flips the encoder to h264_nvenc if you enable it, and if that encode fails it quietly falls back to CPU libx264. So enabling it is harmless, but on a machine where your ffmpeg wasn't built with NVENC support you'll just get the CPU path.

The inputs

• video_path_1 and video_path_2 - required, both plain STRING paths. The order you enter them is the order they get concatenated, so list them the way you want the output.
• video_path_3 … video_path_10 - optional. Empty ones are skipped, so you don't have to fill every slot.
• output_filename - default merged_video.mp4, written to ComfyUI's output directory.
• use_gpu - boolean, default off, the NVENC attempt described above.

One output: merged_video_path (STRING). Feed it into the pack's VideoBasicVideoSave node to register it in the output gallery, or use it downstream as a path.

Install

Same as the rest of the pack - ComfyUI Manager → search ComfyUI-VideoBasic, or:

cd ComfyUI/custom_nodes
git clone https://github.com/jax-explorer/ComfyUI-VideoBasic

then restart. This node is where the pack's deps stop being self-contained, though: it shells out to a system ffmpeg binary, which is not installed by requirements.txt (that file only pulls opencv-python). If ffmpeg isn't on your PATH, the node throws "FFmpeg not found." On Windows that usually means adding it manually or via a package manager; on Linux/macOS it's often already there.

Troubleshooting

• "FFmpeg not found" - it's not a pip problem. Install ffmpeg on the system and make sure it's reachable on PATH, then restart ComfyUI.
• Concat fails when clips lack audio. The concat filter maps [i:a] for every input, so a video with no audio track - like one straight out of the pack's own upscale node, which drops audio - makes it bail to the stream-copy fallback, which then fails unless codecs match. Mixing an upscaled (audio-less) clip with normal clips is the most common way to trip this.
• Fallback needs identical clips. The -c copy file-list path assumes same codec, resolution and fps across everything. If the clips genuinely differ, keep them in the concat-filter path (re-encode) and give every input an audio track.

It's not glamorous, but for "make one video out of these clips" it's the fastest node in the pack - because the only thing slower than FFmpeg at this job is doing it in Python.