🎞️ Split Video

Split Video takes a video file and breaks it into two things ComfyUI can actually work with: a batch of image frames, and the extracted audio track saved to disk. It's the on-ramp for the captioning workflow - you split the clip, send the frames on as a background, and hand the audio to a transcription node.

Why you'd reach for it

ComfyUI thinks in frames and latents, not .mp4 files. To do anything to a video - restyle it, overlay captions, run it through AnimateDiff-adjacent tricks - you first have to turn it into frames. Split Video does that, and crucially it also pulls the audio out in the same step, which is exactly what you need when the next node is going to transcribe that audio into subtitles. One node, both halves.

The inputs that matter

• video - the source clip. This is a dropdown that lists videos available to ComfyUI (typically from your input folder), so you pick a file rather than type a path.
• frame_limit - default 16. The maximum number of frames to pull. This is your safety valve: a 60-second 30fps clip is 1,800 frames, and extracting all of them will eat memory fast. Start small.
• frame_start - default 0. Which frame to begin at, so you can grab a slice from the middle instead of always the top.
• filename_prefix - where the extracted audio file is written, relative to ComfyUI/output (default audio\audio).

Outputs, and there are several useful ones: images (the extracted frames as an IMAGE batch), audio_file (the path to the saved audio, a string you feed into Speech Recognition), fps (the clip's frame rate - pass this to your caption node so timing matches), frame_count (how many frames actually came out), and height / width (the frame dimensions, handy for setting a matching canvas).

Installing it

Ships with Mana Nodes. ComfyUI Manager → search Mana Nodes → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ForeignGods/ComfyUI-Mana-Nodes.git

Install the requirements and restart. Video decode and audio extraction lean on FFmpeg, so make sure FFmpeg is available on your system - on most ComfyUI setups it already is, but a "file won't open / no audio extracted" failure often traces back to a missing or broken FFmpeg.

Common issues

The classic mistake is leaving frame_limit high on a long clip. Extracting thousands of frames into a single batch will balloon your RAM/VRAM and can hard-stop the run. Keep the limit modest and use frame_start to walk through a long video in chunks.

Second, wire the fps output through to your transcription and caption nodes. The whole caption pipeline times words by frame number, so the fps the video was split at has to be the fps everything downstream uses - passing this output along instead of typing a guess is the reliable way to keep captions in sync. Third, if the video dropdown is empty, ComfyUI isn't seeing any files where it expects them; drop your clip in the input folder and refresh. And if audio extraction silently produces nothing, suspect FFmpeg or a video with no audio track.