Video Info (Loaded) 🎥🅥🅗🅢

This is the trimmed-down sibling of the full Video Info node. Where that one gives you ten outputs - the video's properties both before and after your load settings - this one gives you just the five that describe what you actually loaded: fps, frame count, duration, width, height, all post-processing. If you don't care about the original file's numbers and just want a clean node with fewer ports to untangle, this is the one to grab. It's from Kosinkadink's VideoHelperSuite.

Ninety percent of the time, "what did I load" is the only question you're asking. You resampled the video, you thinned the frames, and now you need the real frame rate to keep audio in sync or the real frame count to size something downstream. This node hands you those without the source-vs-loaded outputs you'd just leave unconnected anyway.

How it works

A Load Video node emits a VHS_VIDEOINFO object holding the metadata. This node unpacks only the "loaded" half - the values after force_rate, select_every_nth, and frame_load_cap have been applied. Same data as the blue outputs on the full Video Info node, just without the source (yellow) set.

The input and outputs that matter

One input, video_info, from a Load Video node's matching output. The outputs are the five post-load values: fps, frame_count, duration, width, and height - the width and height being the image-space dimensions even if you loaded straight to latents.

The high-value wire is fps into Video Combine's frame_rate (convert the widget to an input first). That keeps your rendered video playing at the true rate of the frames you loaded, which is what stops audio from sliding out of sync.

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.

Common issues & troubleshooting

I need the original file's fps or frame count too. Then this isn't the node - use the full Video Info, which exposes both the source and loaded sets. This one deliberately drops the source outputs.

The numbers surprise me. They're post-load. If you set force_rate to 8 and select_every_nth to 2, the fps here reflects both - that's the point. Compare against the full Video Info's source outputs if you want to see what changed.

Nothing connects. It needs a Load Video node upstream feeding video_info; it can't read a file on its own.

Width and height look off for a latent workflow. The width and height outputs are reported in image space - the pixel dimensions of the corresponding frame - even when you loaded straight to latents. That's deliberate, so the numbers line up with resize and combine nodes downstream; don't divide them by 8 expecting latent dimensions.