MODEL IDtopazlabs:wonder@3.5

live

Topaz Labs Wonder 3.5

by Topaz LabsAugust 11, 2026

Wonder 3.5 is Topaz Labs' updated generative image enhancement and upscaling model for recovering detail and realism from both high-quality and degraded source images. It builds on Wonder 3 with finer detail generation, fewer repetitive patterns, improved handling of visual noise, and better precision on structured content such as tables, charts, and text-heavy graphics. It is well suited to photo restoration, compressed-image cleanup, creative enlargement, and enhancement workflows that need stronger texture recovery without drifting into overly artificial output.

Controlling detail with enhancement strength

How to control how much detail Topaz Wonder 3.5 rebuilds with settings.enhancementStrength: what low, medium, and high change, and when to dial it down for faces or up for texture.

Upscaling and detail are two separate decisions. upscaleFactor sets how much bigger the image gets, and settings.enhancementStrength sets how hard Wonder works to rebuild detail as it scales. It takes low, medium, or high, and defaults to high, so every upscale is already reconstructing as much texture as it can unless you tell it otherwise.

Step through the same pizza from its soft source to each strength, and the crust and cheese gain texture at every level:

A soft, low-resolution overhead photo of a rustic pizza on a slate board

The pizza upscaled at low strength, cleaner but still restrained

The pizza upscaled at medium strength, with more defined crust and cheese

The pizza upscaled at high strength, with crisp blistered crust and stringy cheese texture

The move from low to high is not sharpening the same pixels harder, it is inventing more texture: the blister on the crust, the pull of the cheese, the flecks of char. A 100% crop of the crust shows how much further high goes:

A close crop of the pizza crust at low strength, smooth and restrained — Low · 100% crop

The same crust crop at high strength, with crisp blistering and char detail — High · 100% crop

The request

Enhancement strength lives under settings, alongside the upscale itself. Set it explicitly when you want a level other than the high default.

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'topazlabs:wonder@3.5',
  upscaleFactor: 3,
  settings: {
    enhancementStrength: 'medium'
  },
  inputs: {
    image: 'https://im.runware.ai/image/os/a14d18/ws/2/ii/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.jpg'
  }
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "topazlabs:wonder@3.5",
            "upscaleFactor": 3,
            "settings": {
                "enhancementStrength": "medium"
            },
            "inputs": {
                "image": "https://im.runware.ai/image/os/a14d18/ws/2/ii/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.jpg"
            }
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "upscale",
      "taskUUID": "c8d9e0f1-2a3b-4c4d-9e5f-6a7b8c9d0e1f",
      "model": "topazlabs:wonder@3.5",
      "upscaleFactor": 3,
      "settings": {
        "enhancementStrength": "medium"
      },
      "inputs": {
        "image": "https://im.runware.ai/image/os/a14d18/ws/2/ii/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.jpg"
      }
    }
  ]'

runware run topazlabs:wonder@3.5 \
  upscaleFactor=3 \
  settings.enhancementStrength=medium \
  inputs.image=https://im.runware.ai/image/os/a14d18/ws/2/ii/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.jpg

{
  "taskType": "upscale",
  "taskUUID": "c8d9e0f1-2a3b-4c4d-9e5f-6a7b8c9d0e1f",
  "model": "topazlabs:wonder@3.5",
  "upscaleFactor": 3,
  "settings": {
    "enhancementStrength": "medium"
  },
  "inputs": {
    "image": "https://im.runware.ai/image/os/a14d18/ws/2/ii/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.jpg"
  }
}

Response

[
  {
    "taskType": "upscale",
    "taskUUID": "c8d9e0f1-2a3b-4c4d-9e5f-6a7b8c9d0e1f",
    "imageUUID": "5b6c7d8e-9f0a-4b1c-8d2e-3f4a5b6c7d8e",
    "imageURL": "https://im.runware.ai/image/os/a14d18/ws/2/ii/5b6c7d8e-9f0a-4b1c-8d2e-3f4a5b6c7d8e.jpg"
  }
]

Reading the three levels

The three settings sit on a faithful-to-invented axis, not a quality dial where higher is always better.

low stays close to the source. It cleans and firms up what is there with the least invention, which is the safest choice when the source is already decent or when the subject has to stay true to the original.
medium is the balanced middle. It rebuilds real detail while holding back on the textures a model is most likely to get wrong, and it is the sensible default for people and mixed scenes.
high reconstructs the most. It produces the richest texture and the sharpest result, and it is also where Wonder is most likely to over-invent on a subject that cannot support it.

Faces need a lighter touch

Skin is the classic place high goes wrong. Pushed hard, it can turn pores and fine lines into an over-etched, waxy surface that reads as processed rather than real. The same portrait at each level shows it:

A soft, low-resolution portrait of a woman with dark hair tied back

The portrait at low strength, clean and natural but soft

The portrait at medium strength, natural skin with recovered detail

The portrait at high strength, skin texture pushed hard and slightly over-etched

Compared side by side at 100%, medium keeps the skin believable while high starts to over-render every pore:

A close crop of the cheek and eye at medium strength, natural skin — Medium · 100% crop

The same crop at high strength, skin texture over-etched — High · 100% crop

For portraits and close-up faces, start at medium and only move to high if the result looks soft. The high default is tuned for detail-rich scenes, and skin is the subject most likely to look artificial when it is pushed too far.

Texture rewards a heavier hand

The opposite is true for subjects that are all fine detail. Fur, foliage, fabric, and food give high real structure to rebuild, so it lands crisp instead of artificial. The retriever's coat is exactly the case high was made for:

A soft, low-resolution close-up of a golden retriever

The retriever at low strength, clean but soft fur

The retriever at medium strength, more defined fur

The retriever at high strength, crisp individual fur strands

Individual strands separate out at high where low leaves the coat smooth. When the subject is texture, the aggressive default is the right one, and dialing down mostly costs you detail.

Tips

Match strength to the subject, not the resolution. Texture-heavy scenes take high well, while faces, flat colour, and simple graphics usually look more natural at medium or low.
Keep the aggressive default for texture. high is the shipped default and the right call for landscapes, fur, fabric, and food. Reach for a lower level when the subject is a person or something that must stay faithful.
Start portraits lower. Skin is the first thing to look artificial under high, so begin at medium and step up only if the face comes back too soft.
Read the result at 100%. The gap between levels is easy to miss at page width and obvious in a crop. Judge skin and fine texture at native size before committing.
Lower strength to stay faithful. When the goal is a clean, honest enlargement rather than a reconstruction, low invents the least and drifts from the original the least.