Very insteresting: Gigapixel AI is the winner, but with reservations!
quoted from the article, in the final conclusions:
"Nor is Topaz Gigapixel AI completely perfect. While I definitely preferred its results over those of its rivals in terms of crispness and detail, it does have an all-too-frequent tendency to miss by some way on color or exposure, and there’s no way to correct this in-app.
You can, however, render to a DNG or just use Gigapixel AI as a plugin instead, and then fix any color/exposure issues in the other app. Hence, I don’t see this as a huge issue, even if I think Topaz Labs definitely needs to look into it.
I’d also like to see the company add both the ability to disable face recovery on a per-face basis and some basic white balance/exposure controls in standalone mode at the least, even if not also when functioning as a plugin."
Unfortunately there are a lot of advantages that Topaz offers that were not sufficiently tested or tested at all. From faces reconstruction, creative effects for illustrations, way it renders different images based on models, like paintings, vs buildings, vs hair, vs grass etc. Vector graphics. Words and letters. Way it deals with heavily compressed JPEG’s etc. All that was not really covered in any detail or at all in the DP review tests. So I really don’t consider it a proper test as much as curiosity. Also there are other products on the market for up scaling that were not tested at all.
Okay, but the negative aspects listed in the conclusion remain clear for all to see:
the handling of RAW files in Gigapixel (but also in Photo AI, Denoise AI and Sharpen AI) at the level of “reading” and processing leaves a lot to be desired, in this the other photo editing programs do much better (precisely in the handling of the “rendering” of the file, from white balance, to colors, to correction of distortion, etc.).
In these aspects Topaz still needs to improve a lot!
Yeah. The only problem off course with lower res files tends to be in certain situations. GFX 100s for example does not have moire problems like some of the full frame mirrorless cameras do, and if you have that in your source image, sometimes Gigapixel does get rid of it, and sometimes it does not. Also it comes down the original RAW file processed. By far I think the best tool right now is DXO PhotoLab 6 with their DeepPrime XD technology of de noising and demosicing with AI. I get much better results even from APS-C cameras than anything before it. Once I have that, than I try to upscale with Gigapixel and that usually produces best results because it gives cleanest most detail file for Gigapixel to work with.
Another problem I found with upscaling is that it works well when the AI model in Gigapixel is trained for certian type of detials. Like brick wall, grass, hair etc. But if you shoot for example texture like sand, as in sand paintings, it fails, mainly because there is no sand paining in its database for training. So it interprets sand texture as something else. This is where original capture with something like GFX would be superior.
That said, for more and more use cases I agree with you. 24-50-61 MP upscaled 2x produces excellent results in most cases.
I wish Fuji GFX 100s would offer what some full frame high res cameras offer now. Ability to shoot with lower res RAW files. Nikon and I think Canon and Sony offer that too. S,M,L , lossless and lossy compressed RAW’s. For all case scenarios. Very useful and versatile. Safes time and disk space in post.
“the handling of RAW files in Gigapixel (but also in Photo AI, Denoise AI and Sharpen AI) at the level of “reading” and processing leaves a lot to be desired, in this the other photo editing programs do much better (precisely in the handling of the “rendering” of the file, from white balance, to colors, to correction of distortion, etc.).”
Definately. I agree. I have noticed that in Video and Photo TopazLabs and color management are not on speaking terms. In their video enhance AI app, they always change colors and in photo apps, they retain appearance usually when working with JPEG or TIFF but in raw its practically unusable.
You know its counter intuitive but with XD I actually crank it up to 70 or 100 %, because it does not just de noise it also demoseics, producing often more details. Even at base ISO, like ISO 100. There is another slider for choosing how much it favors noise vs texture, removal so one can tweak it and also lens sharpness sliders play a major role in what the final result will be like. Perahps worth playing around with.
A small but overlooked tip? Every panel in DXO has a very useful little “?” question-mark in the corner, and when you click on it, it gives you explanation for every tool and slider. Really useful. I’ve discovered some neat little helpful things like that.
Also consider lowering the “bokeh” slider in lens sharpness panel when you boost the DeepPrime XD. You can also use the little “?” in lens sharpness panel to give you info on that. Basically bokeh adds softness across the entire image, while the other two sliders are calibrated to the specific lens sharpness. I found that reducing bokeh slider to 0, I can get nice sharpness that looks natural, and than I need less of other sliders. But depending on the sensor and lens you might have to play around with the sliders to find the sweet spot for your lens. All I can say is that results can be spectacular, best I’ve seen yet, if you get the right settings.
Lateral chromatic aberration (e.g., magenta or green fringes along edges) is automatically corrected only if the appropriate DxO Optics Module is available. In this case, no further manual action is necessary.
You can correct the other types of chromatic aberrations (longitudinal or other) using the two sliders in their respective sections of the palette:
Intensity sets the strength of the correction, within a range of 0 to 200.
Size: adjusts the width of the colored fringe to be suppressed, within a range from 0 to 12 in arbitrary units. This setting affects how DxO PhotoLab determines the chromatic aberration to be corrected, and what is real image content.
You should check the Purple fringing correction box for all backlit scenes, or when shooting with a lens prone to this optical defect.