Load Video FFmpeg (Path) 🎥🅥🅗🅢

This is the newer, faster cousin of the regular Load Video (Path) node. Same job - turn a video file at an arbitrary path into frames - but it decodes with ffmpeg instead of OpenCV, and that swap buys you three real things: it's quicker, it can read transparency (an alpha/mask channel), and it lets you jump to a start timestamp in seconds instead of counting frames. It ships in Kosinkadink's VideoHelperSuite, the standard video-I/O pack for ComfyUI.

There's a specific reason this node has a following. When people chain Wan clips to make longer videos - generate 5 seconds, grab the last frame, generate the next 5 from it - the FFmpeg loader is the recommended way to extract that last frame, because it preserves color far better than the OpenCV path. Fewer generations of color drift across a stitched sequence. If you're doing frame-continuation work, this is the loader you want.

How it works

ffmpeg is the industry-standard media decoder, and using it directly means VHS inherits its speed and its broad format/codec support, including videos that the OpenCV path chokes on or that won't preview in a browser. The alpha channel comes out as a separate mask output, which is genuinely useful when your source has transparency you want to keep.

The inputs and outputs that matter

The video widget is the path string. The knobs you'll actually touch:

• start_time - where to begin, in seconds (this replaces the frame-counting skip_first_frames of the OpenCV node). Want the clip from 0:12? Type 12.
• frame_load_cap - max frames to load, i.e. batch size. 0 means all; set a real cap so you don't exhaust memory.
• force_rate - resample to a target fps to match your motion model (0 leaves it alone).
• custom_width / custom_height - resize on load, 0 to disable.

Optional format picks a model preset (None, AnimateDiff, Mochi, LTXV, Hunyuan, Cosmos, Wan) and adjusts defaults to suit it; an optional vae makes it emit latents instead of images to save RAM on long clips; and meta_batch connects it to a Meta Batch Manager for very long videos.

Outputs: IMAGE (frames), mask (transparency), audio (feed Video Combine to keep sound), and video_info (fps/duration for a Video Info node).

How to install it

ComfyUI Manager: search ComfyUI-VideoHelperSuite, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite

then restart. No model downloads. This node in particular wants ffmpeg present - the imageio-ffmpeg dependency (installed automatically) bundles a binary that covers most cases, but if you hit format errors, install a real ffmpeg and put it on your PATH.

Common issues & troubleshooting

"ffmpeg not found" or a decode error. The bundled binary didn't take, or your file needs a codec it doesn't carry. Install system ffmpeg and make sure it's on PATH, then restart ComfyUI.

The mask output is empty. Your source has no alpha channel - that's normal for most mp4s. Transparency only exists if the file actually stored it (some webm/mov/gif).

Path won't load on a cloud box. If VHS_STRICT_PATHS is set, loading is restricted to ComfyUI's directories; move the file under ComfyUI or use the upload-button version of this node.

Preview ignores my start_time. Enable "VHS Advanced Previews" in the settings gear so the preview reflects the load settings.