Load Video FFmpeg (Upload) 🎥🅥🅗🅢

This is the FFmpeg video loader with an upload button. Same engine as Load Video FFmpeg (Path) - decodes with ffmpeg for speed, transparency support, and second-based seeking - but instead of typing a filesystem path, you pick from a dropdown of files in ComfyUI's input folder and get an upload button to drop new ones in. If you're working in the browser and just want to throw a clip at your graph, this is the one you grab. It's from Kosinkadink's VideoHelperSuite, the go-to video pack for ComfyUI.

Why choose the FFmpeg loader over the plain OpenCV Load Video? Speed, the alpha/mask channel, and - the reason it gets recommended in Wan and vid2vid threads - cleaner color when you're extracting frames to continue a video. Chaining clips (generate, grab the last frame, generate again) degrades color less through this path than through OpenCV. For upload-and-go convenience with that quality, this node is the sweet spot.

How it works

ffmpeg does the decoding, so you inherit broad codec support and the ability to read videos a browser can't natively play. Uploaded files land in ComfyUI's input directory and show up in the video dropdown. Transparency, if the source has it, comes out as a separate mask.

The inputs and outputs that matter

video is the dropdown of uploaded files, with a "choose video to upload" button beside it. The widgets worth setting:

• start_time - begin at a timestamp in seconds. Skip the first 3 seconds by typing 3.
• frame_load_cap - max frames loaded (your batch size). Leave it at 0 and a long clip will try to load in full; set a real number.
• force_rate - resample to a target fps so it matches your motion model's training rate (0 = untouched).
• custom_width / custom_height - resize on the way in, 0 disables.

Optional format applies a per-model preset (None, AnimateDiff, Mochi, LTXV, Hunyuan, Cosmos, Wan); an optional vae switches the output to latents to cut RAM on long sequences; meta_batch hooks it to a Meta Batch Manager for very long inputs.

Outputs: IMAGE, mask, audio, and video_info. Route audio into Video Combine to keep the soundtrack and video_info into a Video Info node when you need the source fps.

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 weights to fetch. The imageio-ffmpeg dependency installs a bundled ffmpeg automatically; a system ffmpeg on PATH is the fallback if you hit a format it can't handle.

Common issues & troubleshooting

Upload succeeds but the file won't decode. Usually a codec the bundled ffmpeg doesn't ship. Install system ffmpeg, put it on PATH, restart.

Empty mask. The source has no alpha - expected for standard mp4s.

The dropdown doesn't show my new file. Refresh the node or reload the page after uploading; ComfyUI rescans the input folder on refresh.

Preview looks wrong / ignores settings. Turn on "VHS Advanced Previews" in the settings gear so the preview honors start_time, frame_load_cap, and force_rate.