Fusion Regions Preview

Once you add regions to a fusion - drawn on a Fusion Studio stage, or grounded by the pack's Qwen3-VL Ground node - the encode can compose its edit from named parts of each reference instead of a blind geometric pattern. But regions are invisible data riding along on the fusion_input. Fusion Regions Preview is the node that makes them visible: it takes a fusion_input, draws each source's region boxes and labels right onto the image, and hands you a preview.

How to read it

It's a pure inspect/debug node in the same spirit as Fusion Inspector, and it's dead simple to use: fusion_input in, preview out, and a Preview Image on the output. One image comes back per source, each at its own native size, so you can flip through what the encoder sees for every reference.

The reading rules matter, and they're the part most people trip on:

• A source with regions gets its boxes and labels drawn on it - that's what the grounding model (or your studio layout) decided to take from that image.
• A source with no region is dimmed to 55% black. That's the author's way of telling you "this image has no regions, so it blends as a full-frame background." It's not a bug - it's the honest picture of what a passthrough looks like in the blend.

The boxes are drawn in each source's own 0..1 coordinate frame - the same frame the grounding boxes were defined in - so the overlay is truthful about what the encode will actually pick up. If a box sits too small or in the wrong spot, you'll see it here before you burn a generation on it.

The one thing it won't do

It does not alter the fusion_input. The output preview is purely a visualization - if you wire this in the middle of your graph expecting regions to be modified, you'll get nothing different downstream. Put it on a side branch off the fusion_input you care about, and leave your main wire alone.

This is the kind of node you add when region fusion (region_strength > 0 on the encode) isn't behaving. The usual failure: a grounding phrase misses, the model returns a degenerate box, and suddenly half the frame is owned by a source you thought was a background. One glance at this preview tells you which image failed and what it found - usually a box too small, too large, or pinned to a corner.

Installing it

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

or search "Nynxz" in ComfyUI Manager, then restart. No dependencies of its own; it just reads the fusion_input structure, so it costs nothing to have around.