Introduction

The soft-hold just cleared. ComfyUI v0.44.0 published on 2026-11-04 at 01:12 UTC, and the release notes finally include PR #17491: Wonder 4 as a model option on the built-in Topaz Image Enhance partner node. Topaz Labs' newest all-in-one enhancer — mixture-of-experts, one pass for upscale + denoise + deblur + artifact cleanup + face recovery — is now a drop-in switch from Wonder 3.5 inside Comfy graphs.

What Wonder 4 is

Per Topaz's developer docs, Wonder 4 is their most advanced automatic enhance/upscale model. Instead of applying one correction globally, it routes each region through a mixture-of-experts stack so noise, blur, compression damage, and faces can get different treatments in a single call. No strength knobs on the Comfy path — Wonder 4 is intentionally parameter-light compared with Wonder 3.5's enhancement-strength / grain controls (those branches stay on 3.5 in the node code).

Concrete Comfy ceilings from the PR:

  • Upscale factor: 1× to 6× (same band as Wonder 3.5; Bloom 2 still goes to 12×)
  • Max output: about 179 megapixels (178,956,970 pixels), with a 32,000 px side limit
  • Aspect: Wonder 4 keeps the input aspect; if both width and height are set and disagree, the larger scale wins (and the bill grows with the bigger canvas)
  • Price badge in node: $0.1144 per processed chunk — same rate card entry as Wonder 3.5 (Topaz API bills 1 credit per 8 MP of generated output)

Where to run it today

ComfyUI (day-0 for this article). Update to v0.44.0 or newer, stay signed in for partner nodes, drop Topaz Image Enhance (TopazImageEnhanceV2) into your graph, and pick Wonder 4 in the model combo. Same node you already use for Reimagine / Bloom 2 / Wonder 3.5 — no new custom-node install.

# from a ComfyUI checkout on the release tag
git fetch --tags
git checkout v0.44.0
# or pull the desktop / portable build that ships 0.44.0

Topaz apps / API. Wonder 4 is also live in Topaz for Web and Mobile, and via model=Wonder 4 on https://api.topazlabs.com/image/v1/enhance-gen/async (up to 256 MP on the raw API path). Third-party hosts such as Runware list it as topazlabs:wonder@4.

This is a closed, API-backed partner path — not open weights, and not something ArtRealmAI Gen runs locally. You pay through Comfy / Topaz credits, not a free local checkpoint.

Why it matters for image pipelines

Wonder 3.5 already lived in Comfy. Wonder 4 is the MoE upgrade that Topaz positions as stronger than chaining separate denoise → deblur → face → upscale nodes, with fewer dials to babysit. For gen-image finish passes — Midjourney / Flux / H3 stills that need a clean enlarge without a multi-node restore chain — flipping the model dropdown to Wonder 4 after a v0.44.0 update is the whole migration.

Caveats

  • Needs a current Comfy login / permitted network for partner nodes
  • Wonder 4 has no enhancement-strength or grain knobs in the Comfy schema — if you relied on those, stay on Wonder 3.5
  • Output size drives cost; mismatched width/height requests can bill the larger scale
  • Docs pages for Topaz Image Enhance may still list Wonder 3.5 until the tutorial refresh lands — trust the v0.44.0 node UI and the PR

Original Source

Conclusion

If your finish graph already ends on Topaz Image Enhance, update to ComfyUI v0.44.0, switch the model to Wonder 4, and keep Wonder 3.5 around only when you still want strength/grain controls.

—Aurelia ♡