Topaz Hyperion HDR Params

Topaz's Hyperion model (hyp-1) is the one that takes standard-dynamic-range footage and produces a genuine HDR result - expanding the highlights and rebalancing colour so the output holds real highlight range instead of a crushed fake-HDR look. Topaz Hyperion HDR Params is the small settings node that feeds Hyperion its SDR→HDR parameters. On its own it does nothing; it produces a params bundle you wire into the params input of a Topaz Video Upscale node whose model is set to Hyperion. If you've got an SDR clip you want on an HDR timeline, that pair is the whole trick.

The three inputs

That's the entire interface, which makes this one of the friendlier nodes in the pack:

• sdr_ip - the SDR highlight threshold (0.45–0.85, default 0.65). It marks where in the SDR range highlights begin, i.e. what the model treats as bright and pushes into HDR headroom. Raise it if you want a subtler lift; lower it to expand more of the midtones.
• hdr_ip_adjust - the exposure adjustment amount (0–1, default 0.5). How hard the model pushes overall exposure up to make room for the HDR range.
• saturate - the saturation adjustment amount (0–1, default 0.5). Rebuilds colour after the exposure move.

The defaults are sensible starting points; think of the three as threshold, exposure, and colour. If your result looks overblown, drop hdr_ip_adjust; if colours look thin after the expansion, lift saturate. There's no "strength" master knob here - the middle slider is the closest thing to one.

Wiring it

The node's one output is params of type TOPAZ_UPSCALE_PARAMS - the same socket type used by Topaz Upscale Params and Topaz Parameter Estimate, which is the giveaway that these three sliders are just another way of filling that bundle. Connect params into the upscale node of your choice and set its model to Hyperion. Same rule as every settings node in this pack: it only affects the processing node it's wired into, so it doesn't touch any other upscale in the graph.

One gotcha to flag before you build expectations: the newer Hyperion-2 isn't reachable here at all. Hyperion-2 (like Astra and Starlight SLP-2.5) runs inside Topaz's neuroserver runtime rather than through ffmpeg, so it can't be driven from outside the Topaz application - the pack hides it rather than offer a model that fails mid-render. hyp-1 is the ffmpeg-reachable one, and this node is tuned for it.

Install

It ships with the pack:

git clone https://github.com/donangel85/ComfyUI-TopazVideoLocal.git

from ComfyUI's custom_nodes folder, restart ComfyUI. No pip install beyond numpy; ComfyUI Manager finds it by searching "TopazVideoLocal". The real requirements live under the whole pack: Windows, a licensed signed-in Topaz Video installation, and the hyp-1 weights on disk ([download required] until you enable allow_model_download on the Engine Settings node or run the model once in the Topaz app).

Common issues

If the output looks identical to the input, the model probably isn't Hyperion - the params node doesn't pick the model for you, so a prob-4 upscale will cheerfully ignore HDR tuning intended for hyp-1. And if you can't find Hyperion in the model list at all, check whether your install only carries the neuroserver-era Hyperion-2, which this pack can't reach; you'll need the hyp-1 weights for this node to mean anything.