Topaz Wonder 3.5 Image Upscaler: Price, Limits & How to Use

Topaz Wonder 3.5 is Topaz Labs’ generative upscaler: it enlarges an image 2× or 4× and rebuilds detail, cleans noise and compression artefacts, and sharpens text and charts. On VdoBloom it runs in the Image Upscaler tab and is priced by the size of the result: 4 credits for outputs up to 8 MP, rising to 45 credits at the 100 MP maximum.

Disclosure: VdoBloom is our platform.

A classic upscaler stretches pixels and smooths them. A generative upscaler like Wonder 3.5 looks at a soft or compressed image and draws plausible fine detail back in: skin texture, fabric weave, leaf edges, crisp letters. This guide covers what the model is, exactly how VdoBloom prices it, when to use 2× or 4×, what the three strength levels do, what no upscaler can recover, and how it compares with the Riverflow upscalers on the same tab.

What Topaz Wonder 3.5 is

Wonder 3.5 comes from Topaz Labs, the company behind Topaz Photo and Topaz Video. It was released on August 11, 2026 and is available in Topaz Photo from version 1.7.0 as well as through API providers. As of September 2026, Topaz’s own documentation describes it as giving “more realistic results with improved detail and text clarity, especially for low-resolution or compressed images, while reducing repetitive patterns and visual noise.”

Compared with Wonder 3, the update targets:

• finer, more natural detail with fewer repeated patterns,
• better handling of noise and compression blocks,
• cleaner text, tables, charts and other structured graphics,
• photo restoration of old or degraded pictures.

Topaz recommends it for small to medium, low-resolution, degraded or heavily compressed images. It is not meant for photos that are already sharp and high-resolution; on those it changes little.

What it costs on VdoBloom

You pay per run, based on the output size in megapixels (width × height of the result). Multiply your image’s width and height by the factor, then find the tier:

One megapixel here means 1,000,000 pixels. Worked examples:

• 1024 x 1024 at 2× gives 2048 x 2048 = 4.2 MP, so 4 credits.
• 2048 x 2048 at 2× gives 4096 x 4096 = 16.8 MP. That is just over 16, so it lands in the 24 MP tier: 11 credits.
• 1500 x 1000 at 4× gives 6000 x 4000 = exactly 24 MP, so 11 credits.
• 1024 x 1536 at 4× (a common AI portrait size) gives 4096 x 6144 = 25.2 MP, so 14 credits.

The tab shows the exact price for your image and factor before you press the button, and VdoBloom re-measures the file when it arrives, so what you see is what you pay.

Size limits

• The largest output is 100 MP. At 2× that means your image can be up to 25 MP; at 4×, up to 6.25 MP.
• The largest input is 64 MP.
• If a request would break either limit, it is refused before any credit is taken, and the tab tells you which factor still fits. A 12 MP phone photo (4000 x 3000), for example, works at 2× (48 MP output, 24 credits) but not at 4×.
• If the upscale fails on the provider’s side, the credits are returned.

How to use Wonder 3.5 on VdoBloom

1. Open the Image Upscaler tab.
2. Upload a JPG, PNG or WEBP image.
3. In the model menu, pick Topaz Wonder 3.5.
4. Choose 2× or 4×. A factor that would exceed 100 MP is greyed out.
5. Choose the enhancement strength: Low, Medium or High (High is the default).
6. Check the credit price shown, then press Upscale Image. The result is saved as a JPG in your creations.

In VdoBloom’s integration runs on September 25, 2026, a 900 x 1359 image upscaled at 2× to 1800 x 2718 in about 18 seconds. At 4.9 MP output, that run costs 4 credits. New accounts get 10 free credits, enough for two upscales in the smallest tier.

2× or 4×: which factor to pick

The provider’s guidance is simple: pick the smallest factor that gets you to the size you need. Bigger is not automatically better, because every extra step means more invented detail.

• Use 2× when the image is already close to the size you want and mainly looks soft, noisy or compressed. Typical cases: a 1024-pixel AI image for a blog header, a screenshot of a chart, a social photo that got recompressed.
• Use 4× when the source is genuinely small and must become large: an old 800-pixel scan for a print, a thumbnail you want as a poster, a small logo or product shot.
• Check the need first. A 2048 x 2048 image at 2× is 4096 pixels wide, enough for most screens and an A4 print. Going further mostly costs credits.

Topaz also advises doing the upscale last, after cropping and colour work, and judging the result at 100% zoom instead of a small preview.

Low, Medium or High strength

According to Topaz’s documentation, the strength setting controls how much new detail the model is allowed to create:

• Low stays closest to the original. It can look slightly softer at 4×, but it is the safest for faces you know well, brand logos and documents where accuracy matters more than crispness.
• Medium is the balance: visible clean-up without heavy reconstruction.
• High reconstructs the most detail and removes the most noise. It suits heavily degraded images and large factors, and it is VdoBloom’s default.

If a High result looks over-textured or a face looks slightly different, run it again at Medium or Low. Each run is priced the same for a given size.

What an upscaler cannot recover

Generative upscaling adds believable detail, not true detail. Keep these limits in mind:

• Lost information stays lost. If a face is a few pixels wide, the model will draw a face, but not necessarily the right one. The same goes for distant licence plates or tiny printed text.
• It can invent texture. On badly damaged images the provider warns the model may confidently add texture that was never there. Compare against the original before you use a restored photo as a record.
• Heavy motion blur and bad focus. Upscaling sharpens edges but does not fully undo a shaky or out-of-focus shot. Our guide to fixing blurry AI images covers when to regenerate instead.
• Noisy low-light phone photos. Topaz lists these as a case where Wonder is not recommended.
• Already sharp photos. Little changes, so you pay for size, not quality.

For photos of real people: only upscale images you have the right to use, and never use restoration to fabricate evidence or misrepresent someone.

Wonder 3.5 vs the Riverflow upscalers on VdoBloom

The Image Upscaler tab also offers two Riverflow 2.0 models, which work differently: you choose a target resolution (2K or 4K) instead of a multiplier, and the price is flat.

For typical AI images around 1 to 2 MP, Wonder at 2× costs 4 credits, which is cheaper than either Riverflow option, and it gives you control over how much detail is added. Riverflow is simpler when you just want “make this 4K” without doing the maths. For a broader look at free and paid options, see our upscaler comparison or the AI image upscaler page.

Frequently asked questions

How much does Topaz Wonder 3.5 cost on VdoBloom?

It depends on the output size: 4 credits up to 8 MP, 7 up to 16 MP, 11 up to 24 MP, and so on to 45 credits at 100 MP. A 1024 x 1024 image at 2× costs 4 credits.

Can I upscale AI images for free?

New VdoBloom accounts get 10 free credits, which covers two Wonder upscales of a typical AI image at 2×. After that, upscales are paid in credits.

How long does a Wonder 3.5 upscale take?

A 900 x 1359 image at 2× took about 18 seconds in VdoBloom’s integration runs. Larger outputs take longer.

Why is 4× greyed out for my image?

The output would be larger than 100 MP. At 4× the source must be 6.25 MP or smaller; choose 2×, or resize the image first.

Will Wonder 3.5 change faces?

It can, slightly, especially on small or blurry faces at High strength, because it redraws detail. Use Low or Medium strength when likeness matters most.

Does it work on text and screenshots?

Yes. Better handling of text, tables and charts is one of the stated improvements in Wonder 3.5. Very small or illegible text can still be misread and redrawn incorrectly, so check the result.