Topaz Gigapixel v1.1.0 - v1.1.1

For those interested I just posted a new idea for Gigapixel - please vote!

1 Like

Oh shit, it turns out Wonder v2 was in Bloom; no wonder I couldn’t find it.:sweat_smile:

Did try Bloom Precision when it came out, didn’t had nearly as good results and skipped it instandly.

It is still Bloom Precision according to the servers:

"intent": "Enhance",
"filters": [
{
"mode": "scale",
"grain": "false",
"model": "Bloom Precision",
"scale": 2,
"width": 1880,

1 Like

ok.

Are these the instructions Gigapixel does send?

Then Wonder is Bloom.

And Starlight was Bloom too?

It’s the reply from the servers for a Wonder job. I guess it doesn’t matter what they call it - as long as it does a good job.

2 Likes


R3 Native → 6000px



Downscale Bilinear → 3766px → Upscale Wonder 2 (4x) → 15064px



I took this picture specifically for testing purposes after noticing that the mushroom has holes on its underside.




Is Wonder result accurate to actual appearance?

Almost.

But it’s really not much.
The model is impressive.

A non expert won’t notice that the images have been enlarged by AI.

I’m still testing.

But I think that if I know I’m going to upscale images later, I won’t shoot below 50mm because the detail resolution, even after downsizing, won’t be sufficient to separate the details properly.
Or the details are swallowed up by downsizing.

I haven’t been able to enlarge to 24000 px yet to see what the model does at 4x 6000 px.

And the groundwork is very important.
The model does not forgive poor quality work or poor lenses.




3 Likes

I can’t tell you exactly.

To see this spider like this, I would have to take a magnifying glass with me. Here, I was 1.5 meters away, according to the EXIF data.



2026 Version - If I don’t re-edit the image and use an existing version, it is not as nearly as good.



before the image was reduced in size.

The logic behind reducing the size is logical.

When enlarged, the image appears blurred because it is technically impossible to capture the image in higher quality.

When reduced in size, connected details can be seen more clearly.



It’s like a dream come true.
Now I can combine speed with resolution.
I’ve even changed my mind about which camera to buy.




V-2026 - Background is Hifi V2.




4 Likes

@alexandre.topazlabs

I have now enlarged a panorama, but it is too large to enlarge in one go, so I have cut it into two parts.

One of the two parts is great, but the other is not quite as good.

The poorer part looks washed out and does not have that sharp, detailed pixel feel.



And it fails with a medium format image that I reduced by 50% (15K image); the text is clearly legible in the reduced image but illegible in the enlarged version.

However, the unreadable version was 2x, not 4x; the software does not support 605 megapixels.

I think that if we have the local version, we can do even more.
We can then specifically reduce the size of the images so that the text is still legible and then use the model in 4x.

:open_mouth:

I need to find out the limits and where they are.

I am looking for the point at which the model works best.

1 Like

We recommend that you use 4x for better results with this new model. Thanks for the test

1 Like

11 posts were split to a new topic: Discussion | Wonder Tests on a spcific machine

A post was merged into an existing topic: Resolved | Model Downloads