Spectrum H3 Objective Media Capture Reset

The sequential benchmark node in this pack holds incomplete R/A/B captures in CPU RAM while it waits for the third role to show up. That's by design - bounded to two benchmark IDs and 4 GiB - but it means a triad you abandon halfway through keeps sitting in memory. Spectrum H3 Objective Media Capture Reset is the official way to flush it.

What it does

Run it and it releases the pending sequential captures for either one benchmark or all of them, then returns a one-line summary telling you what was freed (or that there was nothing to free). It's a pure utility node - no media in, no media out, no comparison involved. Its description is "optional helper to release incomplete sequential objective-media analysis captures from CPU RAM," which is exactly as far as the ambition goes.

Inputs and output

Two required inputs:

• benchmark_id - which benchmark to clear when scope is benchmark. Defaults to h3-objective-seed-1, the same default the capture node uses, so the IDs line up out of the box.
• scope - benchmark clears only the named ID; all clears every pending capture.

Output is a single summary STRING, printed to the console as well, like the other research nodes.

When you'll actually need it

Honestly, not that often. Pending captures are bounded to two IDs and 4 GiB, and the pack also clears them for you when you restart ComfyUI. The reset node earns its keep in two situations:

• You messed up a triad - ran role A with the wrong seed or changed compatibility_tag between runs - and the capture node refuses the next role because the metadata doesn't match. That's the node's consistency guard talking: R/A/B for one benchmark must use identical fps, seed, steps, and compatibility tag. Flip reset_before_capture on the capture node instead, or clear the ID here first.
• You want the RAM back without restarting ComfyUI, and you know you're not finishing that benchmark.

One sharp edge: the sequential capture path only keeps incomplete captures pending. A completed triad has already been evaluated, written to reports, and released - this reset can't un-write a report or retract a verdict. If you want to redo a finished benchmark, clear the ID here (or set reset_before_capture=true) and capture fresh roles.

That's the whole tool. It's a small button, but it's the difference between "wait, that triad is stuck in RAM" and moving on.