Topaz Video Enhance AI

Quick answer, because you're probably here after seeing two identical "Topaz Video Enhance AI" nodes in your node menu: TopazVideoEnhancer and KenChenTopazVideoEnhancer are the same node. The pack registers one class, KenChenTopazVideoEnhancer, and then also registers TopazVideoEnhancer as an alias "just in case." Same inputs, same outputs, same ffmpeg-under-the-hood behavior, same need for a licensed Topaz Video AI install. If you search the node list and get two hits, you're not looking at a bug - you're looking at the author hedging on naming. Pick either; they're interchangeable.

What it actually does

It's the local-workhorse node of a pack that's otherwise all cloud API calls. Given a VIDEO input, it shells out to Topaz Video AI's bundled ffmpeg and runs the tvai_up filter with a real Topaz model - up to 31 of them, from the enhancement family (Artemis, Proteus, Gaia, Theia, Iris, Nyx, Rhea, Dione) to the frame-interpolation family (Themis, Aion, Apollo, Chronos). You control scale (2x/4x/Auto), output resolution (up to 8K), denoise, sharpen, deinterlacing, and frame interpolation up to 8x (240fps). Unlike the Sora nodes, this runs on your own GPU and costs nothing per call - beyond the Topaz license itself.

The inputs that matter

• video - required; feed it the pack's generated clip or any file.
• model - default prob-4 (Proteus), with the full task-prefixed list.
• scale_factor / output_resolution - how big you want it; "Auto" keeps the source ratio.
• denoise / sharpen - Off by default, Low/Medium/High when you need them.
• deinterlace - for interlaced sources.
• frame_interpolation - 2x (60fps)/4x (120fps)/8x (240fps); the marquee feature.
• output_format / quality - mp4/mov/avi/mkv and encode level.
• topaz_path - override for Topaz's ffmpeg if it's not in the default Windows path.

Outputs: output_video (VIDEO) and info (STRING).

Install

The ComfyUI side is the standard pack install (ComfyUI Manager → "ComfyUI_Sora", or clone + pip install -r requirements.txt). The hard dependency is external: Topaz Video AI from topazlabs.com, installed, licensed, and preferably at the default path - the node searches for it if not, and you can point it there via topaz_path.

Common issues

• "未找到Topaz Video AI" - Topaz isn't installed/findable. Install it or set topaz_path.
• Duplicate node entries - the alias means both names show up; they map to the same class, so there's no reason to use one over the other. Standardize on whichever you remember.
• Long runtimes - local Topaz processing on 4K/interpolation jobs is heavy; this is a "go make coffee" node, not a "instant" one.
• Upscaling from a bad source - Topaz enhances, it doesn't resurrect. Feed it the cleanest source you have.

If you already own Topaz Video AI and want it inside a ComfyUI pipeline, this is the node - and whichever of the two names you grab, you're grabbing the same thing.