VELVET VICE · H3 RENDER TIMER

Some nodes do real work - loading models, sampling latents, decoding audio. The VELVET VICE · H3 RENDER TIMER does none of that, and it's honest about it: it has a single enabled toggle and not a single output socket. What it gives you is a live, four-number stopwatch pinned to your workflow so you can actually tell whether that turbo LoRA or resolution change made renders faster or just felt faster.

It's a frontend monitor, not a pipeline stage. The JavaScript watches ComfyUI's execution events and tracks four elapsed times in a small panel on the canvas:

• Current - how long the node currently executing has been running.
• H3 Core - cumulative time inside the MiniMax H3 sampling path specifically. This is the number that tells you whether a change to steps, backend (native vs GGUF), or turbo mode moved the needle where it matters.
• Total - the whole run, from start to finish.
• Last Render - what the previous run's total was, so you can compare two configs back-to-back without writing anything down.

The enabled toggle pauses and resumes the counting rather than ripping the panel off the graph, which is handy when you're doing setup fiddling and don't want to pollute your timing.

You don't wire anything to it, and nothing wires out of it - it's an output/display node that just sits there next to the pack's Preflight monitor. If that sounds trivial, that's the point: in a video workflow where a single render is measured in minutes, wall-clock timing is the difference between knowing turbo mode saved you 40% and believing it saved you 40%.

Install

Same one-pack install as everything else in this suite:

cd ComfyUI/custom_nodes
git clone https://github.com/Velvet-Vice/velvet-vice-minimax-h3

Restart ComfyUI, or use Manager and search "velvet-vice-minimax-h3". No pip dependencies, no model files, no setup.

The honest caveats

Treat the numbers as workflow-level timing, not lab-grade profiling. They track what ComfyUI's execution events report, so the "Current" figure includes whatever node is running at that moment - including non-H3 stages if your graph does pre/post work. The genuinely useful pair is H3 Core and Last Render: core isolates the expensive part, and last-render lets you A/B two settings. And one small gotcha: because it's frontend-only, a crashed or hard-interrupted run may leave the timer mid-state - toggle enabled off and back on to reset the display before the next attempt. Not glamorous, but it beats staring at a frozen 47-minute render with no idea if it's actually still working.