Search across models, guides, examples and the rest of the docs.
↑↓ Navigateenter Openesc Close
Dashboard
live
MODEL IDtopazlabs:wonder@4
Try in Playground

Topaz Labs Wonder 4

Topaz Labs
by
Available with Zero Data Retention

Wonder 4 is Topaz Labs' all-purpose generative model for image enhancement and upscaling. It treats each region of an image separately, combining denoising, deblurring, compression artifact removal, face recovery, and upscaling in one pass with no settings to tune. It suits photo restoration, cleanup of degraded or compressed images, and enhancement at the original size or enlarged up to six times, with output up to 256 megapixels.

Topaz Labs Wonder 4

Restoring photos

How to restore damaged photos with Topaz Wonder 4: faded print scans, over-compressed uploads, low-light noise and motion blur, each fixed in one pass.

Introduction

A photo that needs restoring rarely has a single problem. A scanned print from the 1960s carries faded color and film grain on top of soft focus, and the faces in it are only a few dozen pixels wide. Wonder 4 analyzes each region of the frame separately and applies the correction that region needs, so the faces get face recovery while the grain on the stone steps around them is cleaned up, all from one request with no settings.

A faded, grainy 1960s wedding photo of five people standing in a row on church steps, the faces soft
The same wedding photo, sharp and clean, with clear faces and visible lace on the bride's gown
ScanWonder 4
A faded 760 px print scan restored and upscaled three times. Drag to compare.

The faces carry the most weight in a photo like this, and they start with the least detail. A 100% crop of the couple at the center:

In the scan, each face is soft and grainy, with the eyes little more than dark smudges. The output draws eyes and hairlines at a size that holds up in print, and the grain that covered the skin is gone.

That reconstruction is plausible, not recovered. A face a few dozen pixels wide does not hold the exact shape of an eye or a smile, so what Wonder draws is consistent with the blur rather than read out of it. For a family archive that is the point. Where the exact likeness matters, compare a restored face against another photo of the same person before you ship it.

The request is the same as any Wonder 4 upscale, covered with its size limits in upscaling images. Every example below sends the damaged photo in inputs.image and an upscaleFactor, and nothing else.

Over-compressed uploads

Social platforms re-encode every upload, and a photo that has been saved and shared a few times ends up with blocky compression artifacts in flat areas and ringing along the edges. Wonder treats those blocks as damage to remove, not as texture to keep.

A woman in a camel overcoat and wide-leg jeans crossing a city street, with visible compression blocks across the coat and the background
The same photo, clean, with wool texture on the coat, a defined knit collar and crisp sneaker edges
UploadWonder 4
A street-style photo saved at heavy JPEG compression, upscaled two times.

The camel coat took the worst of it, because compression blocks show first in a large area of one color. In the output the coat reads as wool again, and the edges of the jeans and sneakers are clean lines instead of stepped ones.

Low-light noise

A phone in a dim room raises its sensor gain to get an exposure, and the cost is colored speckle across every dark area. Ordinary noise reduction trades that speckle for smear, because a filter cannot tell grain from real texture.

A coach holding focus mitts for a boxer in red gloves in a dim gym, covered in heavy color noise
The same scene, clean, with smooth dark shadows, defined mitts and gloves, and skin texture on the coach's forearms
SnapshotWonder 4
A noisy low-light snapshot denoised and upscaled two times.

Wonder handles the two kinds of region differently. The dark background goes smooth, while the mitts and the coach's forearms keep their texture, so the result does not have the waxy look of a frame that was simply blurred clean.

Motion blur

A camera that moves during the exposure leaves a sideways smear across the whole frame, and a sharpening filter can only outline it. Youth sports are full of it, because parents shoot handheld from the sideline while following the play.

Two children in yellow and blue jerseys chasing a soccer ball on grass with parents watching from the sideline, the whole frame smeared sideways by motion blur
The same moment, sharp, with defined faces, folds in the jerseys and the panels on the ball
Action shotWonder 4
A motion-blurred shot from a youth match, deblurred and upscaled two times.

The blur in the source runs sideways across the players and the crowd alike, and the output resolves both back into edges, down to the panels on the ball.

Tips

  1. Send the original file. Every re-save adds compression the model has to remove, and the detail under it is gone for good. A direct scan or the camera export gives Wonder more to rebuild from than a copy pulled off a social feed.

  2. Pick the factor from the size you need. Restoration runs at every factor, so choose it from how large the result has to be. A photo that is already big enough can be cleaned at its own size, covered in enhancing at the original resolution.

  3. Check faces at 100%. A restored face looks right at page width almost every time. Zoom in before you ship it, and compare it with another photo of the same person when you have one.

  4. Keep the untouched source. The restoration is a new image with rebuilt detail, so keep the original scan or upload as the record and treat the restored file as a derivative.

Ask AI
Send feedback