Topaz Engine Settings

This node doesn't process a single frame. It's the pack's configuration hub - a single set of engine-wide options that any other node in the pack can consume, so you don't repeat yourself across a graph. You don't need it on the canvas for basic runs; every node works with defaults. You reach for Topaz Engine Settings when something specific needs changing: the install wasn't auto-detected, a model's weights need downloading, or you're debugging a failed render and want to see the actual ffmpeg command.

What it controls

Wire its engine output into the engine input of any processing node (upscale, interpolate, stabilize, deinterlace…) to apply these. The ones that matter in practice:

• install_path - point at your Topaz Video folder when auto-discovery fails. A folder only counts if its ffmpeg provides tvai_up. This is the fix for "No usable Topaz Video installation found."
• device - GPU index: -2 auto (default, strongly recommended), -1 CPU, 0 first GPU. The tooltip carries the warning from real testing: on mixed NVIDIA/AMD systems an explicit device=0 has been observed to fail with "Failed to configure output pad". Leave it on auto.
• allow_model_download - let Topaz fetch missing model weights from its servers. Off by default so processing stays fully offline (your footage is never uploaded either way). Turn it on and [download required] models just work.
• transport - pipe (default): raw frames straight into Topaz, no decoder involved. file: a lossless intermediate file instead - slower, but easier to inspect when debugging.
• license_check - cached (verify once and remember across restarts), force (re-verify now), or skip. The licence is checked once and cached in config.json, keyed to the installation.
• verbose - log the exact ffmpeg command line for every call. When a render fails, this gives you the command to paste into a terminal and reproduce.
• vram - fraction of video memory Topaz may use (0.1–1). instances - extra parallel model instances (the tooltip's advice: 0 is usually best).

The tail of the list - keep_temp_on_error, timeout_seconds (abort a run after N seconds; 0 is no limit) - is for specific debugging. And save_as_defaults persists the current values to config.json so they survive ComfyUI restarts, which is how you make an install_path fix permanent rather than per-workflow.

Outputs

One output: engine, of the pack-private TOPAZ_ENGINE type. It doesn't go anywhere except into another node's engine socket, and it only applies to nodes downstream of it. If you want engine settings to apply everywhere, wire it into every processing node (or save as defaults and don't wire anything).

Install

Same as the whole pack:

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

from ComfyUI's custom_nodes folder, then restart. No pip install beyond the numpy ComfyUI already has; ComfyUI Manager finds it under "TopazVideoLocal". And remember the layer underneath the node's own settings: this pack runs on Windows against a licensed, signed-in Topaz Video installation.

Common issues

The two mistakes people make with this node are wiring it wrong and not saving it. A graph with two Engine Settings nodes feeding different processing nodes can silently disagree - use one, or use none. And values you set on the canvas die with the workflow unless save_as_defaults is on; if you fixed install_path once and it's broken again in a fresh workflow, that's why.