Topaz Upscale Parameters

This one's small and it doesn't do any actual video processing - it just packages up an upscale preset. TopazUpscaleParams is the same four upscale sliders you'd otherwise set directly on the main TopazVideoAI node, pulled out into their own node so you can build a setting once and pass it around, instead of retyping the same values into every TopazVideoAI node in a busy graph. It only makes sense paired with that node (or TopazUpscaleParams-consuming nodes more broadly) - on its own it just sits there holding a preset.

Worth flagging up front: the pack's README doesn't mention this node by name at all - every install and usage instruction in it is written for TopazVideoAI singular. So some of what follows about how it's meant to be used is a reasonable read of the field names, not something documented anywhere. Read it as an informed guess, not gospel.

What it does

It takes the same upscale-related fields as the main node - upscale_factor, upscale_model, compression, blend - and bundles them into a single UPSCALE_PARAMS object. That output is exactly the type TopazVideoAI's optional previous_upscale input accepts. So the intended flow looks like: set your upscale preferences once in a TopazUpscaleParams node, then feed that into one or more TopazVideoAI nodes downstream, rather than configuring the same four sliders repeatedly.

The previous_upscale input on this node (not just on TopazVideoAI) is the more interesting detail - it accepts the same UPSCALE_PARAMS type this node outputs, which is a chaining pattern: one TopazUpscaleParams node's output can feed into another's previous_upscale input. What exactly the chain does downstream isn't spelled out anywhere in the pack, but the shape strongly suggests it's meant for building a short list of upscale passes rather than just one flat preset.

The fields that matter

• upscale_factor (1–4, default 2) and upscale_model (16 choices, default iris-3 here - note it defaults differently than TopazVideoAI's inline thf-4). Same caveat applies as on the main node: not every model in the list has been validated at every scale factor, and thm-2 is fixed at 1x. If a high factor throws an error, drop to 2x.
• compression (-1 to 1, default 1) and blend (0–1, default 0) - the same two secondary Topaz sliders exposed on the main node. Leave them at defaults unless you already know what you're chasing.
• previous_upscale (optional) - chains in another params node, as above.

Output is a single UPSCALE_PARAMS - not an image, not a video, just a settings bundle. Wire it into TopazVideoAI's previous_upscale input.

How to install it

It ships in the same pack as TopazVideoAI, so there's no separate install. Search TopazVideoAI in ComfyUI Manager and install, or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/sh570655308/ComfyUI-TopazVideoAI.git

and restart. That gets you the node itself, but it's inert without the rest of the stack: a licensed, logged-in Topaz Video AI app on the same machine, the TVAI_MODEL_DATA_DIR / TVAI_MODEL_DIR environment variables set (Topaz can auto-set these - open the Topaz app, log in, hit Ctrl+Shift+T), and a one-time .\login run from a terminal in Topaz's install directory. None of that is specific to this node, but since TopazUpscaleParams is useless without a TopazVideoAI node consuming its output, you need the whole chain working before this node's presets do anything visible.

Common issues & troubleshooting

Because this node never touches ffmpeg or your video frames directly, almost nothing goes wrong here - the failures show up downstream, on the TopazVideoAI node that actually consumes the preset. If your graph errors out:

Check the consumer, not this node. A No such filter: 'tvai_up' error or a silent failure almost always traces back to TopazVideoAI's ffmpeg path or an un-authorized Topaz login, not to anything set here.

Scale-factor mismatch. If you built a preset at 4x and it errors, that's the same known limitation as the inline sliders - some models aren't validated past 2x. Try 2x before assuming the node is broken.

Nothing seems to apply. Double-check the UPSCALE_PARAMS output is actually wired into previous_upscale on the node you expect it to affect - it's an easy connection to forget since it's an optional input, not the node's main data path.