H3 Extras

The H3 pack keeps its state on a h3_inputs wire, and that's great for plumbing - but a wire is opaque. H3 Extras is the probe that reads the canvas back off it as plain numbers, the ones you can wire into a resolution picker, a text node, or a graph branch.

What you get

Feed it any h3_inputs wire and mode (default auto - match whatever H3 Encode is set to, so the task reported is the task that'll actually run), and it splits the canvas into values:

• width, height - the snapped canvas.
• length - the length the canvas asked for, and frame_count - what it snaps up to on H3's 17k+5 frame grid. These differ, and the difference is the point: ask for 5 seconds and you get 124 frames, not 120.
• latent_t (video latent frames) and audio_t (audio latent frames at 40 Hz) - the actual latent lengths.
• seconds - frame_count / 24, the real duration.
• task - t2va, fl2va or ref2va - what the same list will make H3 Encode do.
• is_keyframes / is_references - booleans you can wire into a switch to branch a graph on which task is active.
• summary - one human-readable line naming all of it, for a text preview.

Why you'd bother

H3 Canvas already gives you a report string with most of this. The difference is type: report is prose for your eyes, H3 Extras is numbers for your graph. is_keyframes into a conditional is the obvious branch - run a different sampler or schedule depending on whether a keyframe is driving the clip. frame_count into a text node is the honest way to show the user what they're actually going to get, and latent_t is what you want if any downstream node needs the true latent length rather than the requested one.

Install

Pack install: ComfyUI Manager → "Nynxz H3", or:

cd ComfyUI/custom_nodes
git clone https://github.com/Nynxz/ComfyUI-NynxzH3

Restart. Pure Python, no dependencies, ComfyUI 0.30.0+.

The one-liner

If you ever find yourself re-typing the frame count a workflow reported, this is the node that hands it to you as a wireable integer. Small, but it's the difference between reading your own logs and letting the graph decide for itself.