Topaz Image Enhance

Topaz is the desktop upscaler everyone knows - the "industry standard" software you either bought a license for or watched other people use. This node is Topaz's hosted enhance API, dropped straight into ComfyUI as a partner node. Give it an image, pick a model, and it comes back bigger and cleaner. It's the easy button for upscaling, and here's the honest take up front: easy, consistent, and fast - but not the quality ceiling. A well-built tiled diffusion + ControlNet Tile workflow can out-fidelity it, as the community will happily tell you. What this buys you is that you never have to build one of those.

What it is. A partner node that sends your image to Topaz's cloud enhance service through Comfy's proxy. Crucially, this is not your local Topaz install talking - there's no desktop app involved. You're billed per call against Comfy account credits, so it behaves more like a metered API than a subscription you own.

How it works. The node uploads your image, kicks off an async enhancement job, polls the status every few seconds, and downloads the finished result. That's the mechanism; from your side it just looks like a node that takes a while. The real variation is in the model dropdown, which is a dynamic combo that changes the controls you get. Three model families:

• Reimagine - creative upscaling guided by an optional text prompt, with a creativity slider from 1 (restrained) to 9 (pronounced reinterpretation). The one you'd reach for when the source is rough and you want it improved, not just enlarged. It also has face enhancement/preservation and subject_detection (All/Foreground/Background).
• Bloom 2 - generative reinterpretation. Takes a creativity slider, a seed, and grain controls. "Upscale with new detail" rather than "faithfully enlarge."
• Wonder 3.5 - the clean, faithful enhancer with an enhancement_strength of low/medium/high. This is the everyday "make it bigger without inventing stuff" option.

The inputs that matter. image and model are the only required ones. output_width / output_height are optional, and the zero-value semantics matter: 0 means "figure it out" (usually original size, or scaled proportionally if you set the other dimension). Note that Bloom 2 and Wonder 3.5 preserve the input aspect ratio and treat your requested size as a target - and Wonder 3.5 only supports upscale factors from 1x to 6x. When the output aspect differs from the input, the result gets letterboxed by default; Reimagine's crop_to_fill toggle changes that to a crop.

The outputs. One IMAGE, ready for the same save/preview/compositing nodes as anything else. Nothing exotic downstream.

How you get it. Built into ComfyUI core, no extension, no model files to download - everything runs in Topaz's cloud. You need a Comfy account with credits and a logged-in session, and your image leaves your machine to get enhanced, so keep that in mind for anything sensitive.

Where people get burned. The main trap is treating it as a drop-in replacement for a custom upscaling pipeline and then being disappointed by fidelity on heavy upscales - it's a fast generalist, not the record holder. Second, the width/height zero semantics confuse everyone once: if you set only output_width, height scales proportionally; set both and you may be asking for a target that aspect-preserving models will quietly adjust. And since these partner nodes landed in the mid-2026 wave, the model list will keep changing - expect new Topaz models to appear as they're released, and keep ComfyUI updated.