Topaz Image Enhance (Legacy)

Upscaling is three different jobs wearing the same name. "Make it bigger" (pixels only, nothing invented), "make it bigger and more detailed" (the model invents texture that was never there), and video. The Topaz Image Enhance node is firmly in the second camp - it's the generative rung, the kind of upscale that adds pores, fabric weave, and sharpness rather than just resampling.

Topaz's name carries weight in this space - it's the photo-enhancement company the whole industry measures against, and this node brings its flagship "Reimagine" model into a Comfy graph. One caveat up front, because it affects how you should use this: the version you'll find in ComfyUI today is marked legacy, replaced by TopazImageEnhanceV2 with a wider model list. It still runs; just know there's a newer door.

What it actually is

A core partner node: built into ComfyUI, no install, cloud-backed. Sign in with your ComfyUI account, spend credits, and the enhancement runs on Topaz's servers via Comfy's proxy. It's synchronous - the node uploads your image, submits the job, polls the status endpoint, and drops a finished IMAGE tensor back into your graph.

The inputs that matter

• model - Reimagine, the only option on the legacy node. It's the generative, prompt-aware upscaler: give it a creative kick and it will invent detail.
• image - exactly one image. Batches error out.
• prompt (optional) - the creative steering. This is the input that separates "Reimagine" from a plain sharpener: describe the look you want the upscale to lean into.
• creativity (1–9, default 3) - how freely the model invents. Low stays conservative; high hallucinates texture.
• subject_detection - All, Foreground, or Background - where the enhancement budget goes.
• face_enhancement (on by default), plus face_enhancement_creativity and face_enhancement_strength - faces get their own pass. Handy, but see the warning below.
• face_preservation / color_preservation (both on by default) - keep identities recognizable and colors true.
• output_width / output_height (0 = auto) and crop_to_fill - if you fix the output dimensions, the result letterboxes by default; flip crop_to_fill to fill instead.

The trap to respect

Every generative upscale rewrites faces - that's the whole community's caution in the KB's upscaling essay, and it applies here harder than anywhere. Reimagine can add detail, but it can also invent a face that only sort of looks like the subject. If identity matters, keep face_preservation on, don't crank creativity past 5, and treat the output as "inspired by the original" rather than a restoration. For objects, texture, and scenes, it's a different story - creative upscaling is where this node shines.

Also worth remembering: generative upscaling is the expensive rung for a reason. If your source is already sharp and you just need more pixels, a local ESRGAN or Lanczos pass is free, instant, and can't hallucinate. Reach for Topaz when you want the detail the source never had - not when you just want a bigger file.

Gotchas

• It costs credits per run, and the price scales with the credits the job reports back - the node tracks it, so glance at the badge before queueing big outputs.
• The legacy status is real: the newer V2 node exposes more models (Reimagine alongside the newer creative upscalers). If you're building a fresh workflow, it's worth checking whether V2 is in your version.
• The usual partner-node quirks: needs a Comfy account with credits, and login works from localhost rather than a LAN --listen setup.

Sharp source, want pixels? Stay local. Soft source, want detail that was never captured - or a creative reinterpretation? This is the node.