FL Load Video

The name undersells this node. FL Load Video doesn't just load a video - it's the front door for the whole video-to-video side of the Fill-Nodes pack. It decodes a file from your ComfyUI input folder into a frame batch, and while it's at it: trims to a time range, resamples the frame rate, resizes, caps the frame count, and hands you the audio track as its own output. One node, and you have everything a video workflow needs before the sampler ever runs.

ComfyUI core still ships no built-in video loader, so this space belongs to community packs (Video Helper Suite is the one most people reach for). FL Load Video is the Fill-Nodes answer, and its edge is the outputs: images, audio, a native VIDEO, plus fps and frame_count as raw numbers. For a video-to-video pass on a Wan or LTX-Video workflow, you feed images into the VAE, sample, and then on the far side you want the original audio back - this node carried it through, so a combine node can mux it in without you hunting down the source file again.

How it works

It's PyAV under the hood: probe the file, decode only the range you asked for, then sample. Every behavior lives in one load_settings JSON blob, which is fiddly at first but means the whole load plan is reproducible and shareable. A built-in memory check estimates peak RAM from frame count, resolution, and FPS, and refuses to load if you're about to blow up your machine - the error message tells you to reduce the range, frame count, FPS, or resolution. Trust it, it's usually right.

The inputs and outputs that matter

video - a dropdown of video files in ComfyUI's input folder (drag a file into ComfyUI/input, or straight onto the node). It accepts avi, gif, m4v, mkv, mov, mp4, and webm - and it must live in the input directory; the node rejects absolute paths pointing elsewhere.

load_settings - the JSON config. The fields you'll actually touch:

• start_time / end_time - trim window in seconds. An end_time of 0 means "run to the end."
• sample_mode - source keeps the native FPS, target_fps resamples to target_fps, every_nth keeps every Nth frame via select_every_nth.
• frame_load_cap - 0 means every frame; set a number to keep long files from flooding your VRAM and RAM.
• resize_mode - original, fit, or crop, paired with width / height.
• include_audio - whether the audio track survives to the output.

The outputs wire up like this: images (the IMAGE frame batch) goes into your VAE or sampler, audio (AUDIO) feeds a video-combine node downstream, video is a native ComfyUI VIDEO that plays in the preview, and fps / frame_count drive batch-size or timing math elsewhere in the graph.

Installing it

The node ships inside the big Fill-Nodes pack, so you're installing the whole thing for these two video nodes - that's fine, it's how it's distributed.

ComfyUI Manager - search "ComfyUI_Fill-Nodes" and install.

Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI_Fill-Nodes
cd ComfyUI_Fill-Nodes
pip install -r requirements.txt

Then restart ComfyUI. The pack is Patreon-funded by the author (filliptm / Machinedelusions) and the requirements list is heavy - librosa, opencv-python, moviepy, google-genai, fal-client, and friends - but you only pay for the ones that import.

Common issues

• ModuleNotFoundError: No module named 'av' - the real gotcha. The video nodes import PyAV, but requirements.txt doesn't list it, so a fresh install can load the pack and still fail here. Fix: pip install av in ComfyUI's Python environment.
• Drag-and-drop stopped working after a newer ComfyUI frontend update. The frontend changed how drop events route, and a community patch restoring image/workflow drops was merged into the main repo in mid-2026 - update the pack and it comes back.
• "Must be inside the ComfyUI input directory" - you dropped a file from elsewhere on disk. Move it into ComfyUI/input and pick it from the dropdown.
• Import errors on very old ComfyUI - the node relies on comfy_api and native VIDEO/AUDIO types that older builds don't ship. Keep ComfyUI updated and this stops being a question.