H3 Studio · Reference Comparison
Making the Benchmark Lab grid actually legible
- images
- studio_context
A raw comparison grid is useful right up until you can't tell which cell is which. H3StudioComparisonView is the pack's presentation layer for that problem: it takes the benchmark output plus the studio context and renders it as a proper comparison sheet - with the source references labeled, the original prompt shown, and the cell metadata attached so "the good one" isn't just a vague memory.
The description calls it an "optional presentation sheet controlled by Director · Comparison image," which is a fair summary of its role. It has two inputs: images (the comparison grid from the Benchmark Lab) and studio_context (from the Director, so it knows what was being compared). It's an output node - OUTPUT_NODE is set, so it displays in the UI and has no downstream outputs to wire.
Why it exists
The Benchmark Lab's raw comparison_grid output is information-dense: a contact sheet of cells with per-cell timing and seed data in the report string. When you're comparing two sampling profiles you need to remember which column was LightX and which was Base, what seed each used, and what resolution each cell actually produced - the grid alone doesn't carry that. This node puts the context back beside the pixels, which is exactly what you want when you're deciding whether a 4-step profile is "good enough" versus the 20-step baseline.
It also plays the same role for the reference comparison sheet: if your benchmark includes references, the view labels them with their correct @ImageN cards so you can see which cells honored which reference.
When you'd use it
If you're running benchmarks at all - comparing profiles, resolutions, or the VAE decode comparison - this node is worth adding at the end of the graph even though it's optional. It costs nothing (no generation, just presentation) and it makes the result reviewable later, which matters when you've queued a 12-cell matrix and walked away for ten minutes. It's also handy for saving a workflow's comparison output to PNG via a save node if you want a record.
The one honest caveat: it only adds value when there's something to present. Wire it after the Benchmark Lab's comparison_grid, keep it out of your normal generation path, and you'll never notice it's there until you need it.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/thaakeno/ComfyUI-MiniMax-H3-Studio.git
cd ComfyUI-MiniMax-H3-Studio
python -m pip install -r requirements.txt
Restart and hard-refresh the frontend. No extra dependencies - it's a pure display node.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | — | |
| studio_context | H3_STUDIO_CONTEXT | — |
Outputs (0)
No outputs