Topaz Video Enhance AI

KenChenTopazVideoEnhancer is the pack's odd duck: the one node that does real work on your own machine. Everything else in ComfyUI_Sora is an API wrapper that phones a reseller, but this one shells out to Topaz Video AI, the paid desktop app, and runs its enhancement models on your GPU locally. If you're generating short clips with the pack's Sora nodes and want them sharpened, denoised, deinterlaced, upscaled, or frame-interpolated before publishing, this is the end of the line. It's also the node with the hardest prerequisite in the pack: Topaz Video AI is commercial software, and it costs real money.

How it works

Under the hood it's a thin but effective wrapper around Topaz's bundled ffmpeg. The node maps a friendly model name to Topaz's internal model ID (Artemis High Quality → ahq-12, Proteus → prob-4, Gaia, Theia, Iris, Nyx, and the frame-interpolation models Themis/Aion/Apollo/Chronos), then builds an ffmpeg command using the tvai_up filter with your settings: scale factor, denoise (mapped to a compression parameter), sharpen, deinterlace, and frame rate. The output lands in your ComfyUI output folder. The model dropdown has 31 entries and they're real - these are the actual Topaz model IDs the desktop app uses, which is the difference between this node working and being a dead wrapper.

The inputs that matter

• video - required VIDEO input; feed it the pack's Sora output or any local clip.
• model - the big list, prefixed by task: Frame Interpolation | … for frame-rate models, Enhancement | … for quality models. Default prob-4 (Proteus).
• scale_factor - 2x, 4x, or Auto.
• output_resolution - Auto, 720p, 1080p, 2K, 4K, 8K. 8K from a 720p source is asking a lot; be realistic.
• denoise / sharpen - Off/Low/Medium/High. Off by default - good, because cranking both makes video look AI-greased.
• deinterlace - for interlaced sources (DV, old TV captures).
• frame_interpolation - Off, 2x (60fps), 4x (120fps), 8x (240fps). This is where Topaz genuinely shines; a 30fps clip to 120fps is buttery.
• output_format / quality - mp4/mov/avi/mkv and encode quality.
• topaz_path - optional override; defaults to Topaz's Windows ffmpeg path, and the node also searches for the install if that's wrong.

Outputs: output_video (VIDEO) and info (STRING - status/errors).

Install

The node itself is the usual ComfyUI Manager → "ComfyUI_Sora" install (or clone + pip install -r requirements.txt). The real install is Topaz Video AI itself from topazlabs.com, at its default path. The node is not an API call and not a bundled model - no Topaz, no node.

Common issues

• "未找到Topaz Video AI" - Topaz isn't installed or isn't at the expected path. Install it (or pass the ffmpeg path in topaz_path).
• Slow - this runs on your GPU locally, and 4K enhancement is minutes-per-clip, not seconds. The community benchmarks against exactly this: a decent local upscaler can beat Topaz for some workflows, but Topaz remains the paid default people compare against.
• Quality surprises - Artemis - High Quality for clean sources, Low Quality/Medium Quality variants for noisy ones. Match the model to your source or you'll trade one artifact for another.

Reach for it when you need the best local upscale/interpolation you can buy and it's already in your toolchain. If you're not a Topaz customer, the pack has no free fallback - the FrameBlender node is the only other local video tool in the box.