H3 Continuum Load Video

Most video loaders in ComfyUI give you frames and shrug about the sound. H3ContinuumLoadVideo is the one built for the awkward fact that MiniMax H3 video comes with audio attached, and Continuum wants to treat that audio separately from the picture. Its two outputs - images and audio - are meant to split down different paths on the H3 Continuum Sampler V3.8: the frames into Video Guide Frames as persistent visual guidance, the audio into Driving Audio only when the source's own soundtrack should guide the clip and survive as the final output. Wire both and your source clip's voice comes out the other end.

Under the hood it's a thin wrapper over ComfyUI Core's VideoFromFile, with the loader's own frame-rate control bolted on. It decodes the file, resamples frames if you ask, and passes the audio through untouched.

The inputs that matter

Three controls, two of which you'll actually touch:

• enable_video - the bypass switch, and unlike the image/audio loaders it's a real input here, not just frontend chrome. OFF returns an execution blocker so the whole optional path behaves as unconnected.
• file - the video you're loading, from ComfyUI's input folder.
• force_rate - the one with a gotcha in its tooltip: 0 uses the source FPS; a positive value drops or duplicates frames to that rate while preserving nominal duration and audio. Default 24, which is deliberate.

That last point is where people get burned. Continuum interprets Video Guide frames at 24 fps, so if your source is 25 fps or 30 fps and you want temporally aligned guidance, set force_rate to 24 (or resample the file before loading). Leave it at 0 for an already-24fps source - don't resample for the sake of it.

Outputs

images (the frame batch, for Video Guide Frames) and audio (for Driving Audio). If the file has no audio track, the audio output is an execution blocker rather than a crash - the graph just proceeds silently.

Installing

It ships in the H3 Continuum pack, so this is the whole-pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum.git

Restart ComfyUI (or find "H3 Continuum" in ComfyUI Manager). Zero extra pip packages; the pack needs ComfyUI 0.32.0+, and the code paths this node uses assume a recent Core with the VideoFromFile API.

Worth knowing before you lean on it

A video guide is exactly that - guidance. It steers motion, framing and appearance; it does not reproduce every source frame, and the pack's own limits section says Continuation doesn't guarantee frame-perfect identity. And remember H3's native generations cap around 15 seconds per pass, which is precisely why this pack chunks - a longer reference clip will be trimmed or padded to what a single chunk can carry. Oh, and as with every node in the pack: the license on the MiniMax H3 weights excludes the US, EU, UK and South Korea, so check yours before building a long-form pipeline on it.