MiniMax H3 Run Summary

If you've ever dug through a folder of renders a week later wondering which seed and which prompt made the good one, the MiniMax H3 Run Summary is for you. It's a deliberately boring node: you tell it what you're about to run, and it hands you a plain-text record of that run. That's it. But when H3 clips are slow and VRAM-hungry to produce, "boring record keeping" is quietly the most useful node in the pack.

This is the one node in the toolkit that's an output node - it's marked as a terminal point in your graph, so its text shows up in ComfyUI's results panel the way a saved image would. That's intentional: the run summary is meant to be the last thing your H3 workflow touches, catching all the settings so you don't have to write them down yourself.

How it works

Give it the settings, get a stable, machine-readable block back:

Mode: T2VA
Resolution: 768p class
Duration: 5s
Seed: 42
References: None recorded
Prompt:
A ceramic robot turns toward a workshop window...

The exact fields are mode, resolution, duration_seconds, seed, references, and prompt. Every one of them ends up as a labeled line in the output, so the summary is copy-pasteable into a notes file, a spreadsheet, or a comment on a shared clip - no JSON, no ComfyUI project file required.

The inputs that matter

seed and duration_seconds are where reproducibility lives. seed ranges up to the full 64-bit space (0 to 18446744073709551615), so you're never truncating the randomness the way some nodes do. duration_seconds runs 1 to 300 - 5 seconds is the sensible default for testing, and you'll thank yourself for logging it when you're comparing a 5s and a 15s take. references is the freeform field for listing whatever reference images, frames, or audio you passed in.

The natural wiring is to feed this node the prompt output of the MiniMax H3 Prompt Builder, so the summary captures the exact prompt that produced the clip instead of a paraphrased version. There's no magic here - no hashing of your workflow, no auto-detection of settings. You set it, it logs it. If you forget to update a field, the summary reflects the lie.

Install

Same story as the rest of this pack: ComfyUI Manager, search MiniMax H3 Workflow Toolkit, or:

cd ComfyUI/custom_nodes
git clone https://github.com/jayseanbrambila-del666/ComfyUI-MiniMax3-H3-Toolkit

Restart, look for "MiniMax H3" in the node library. Zero third-party dependencies, zero model downloads, zero network calls.

Common gotchas

The honest limitation: this node records what you tell it, not what actually ran. If you change the seed in the KSampler but leave 0 in the Run Summary, you've built a misleading log. There's no auto-sync with your sampler - either keep the fields in step or treat the summary as documentation of intent.

And the pack-wide reminder: nothing in this toolkit generates video. The actual MiniMax H3 weights (~42.5GB) and conditioning workflow are a separate setup, and the weights' Community License geofences them out of the US, EU, UK, and South Korea. What this node gives you is the record of the run - the run itself is your job.