[Ian Stuart Forsyth]

Garbage in - garbage out.

One of the things for me when people use some software enhancements for correction with either NR along with sharpening produce blotchy square patterns around the subject with areas in the photo that appear with what looks like detail in the areas of the square patterns that should be nicely soft OOF. More and more I am seeing this while simple in the field adjustment in the camera setting would have worked much better than applying software based correction.

[3by2]

Late to this one, not unusual. I've been trying out the trial. I've often thought that some of the results from the Ai software doesn't look natural and that's still true. My results are a mixed bag but I've had enough good results now to make the hop and buy it. In particular I've been trying it on scanned slides. Some it's failed miserably at, really enhancing the noise detail instead of removing it, others it's restored beautifully. One slide suddenly looked like it did on the slide. It's a slide I've scanned numerous times on a film scanner and by camera. A really high dynamic range and nothing has managed to get it right. I ran it through Gigapixel AI and voila, suddenly I was looking at the slide as I was seeing it held up to the light with a magnifier on it. I couldn't quite believe it. Detail, colours, dynamic range, all spot on. I did a few, what the.... double takes. It's also working pretty well on some older Canon S30 3Mp files.

I shall await one of their offers and purchase it. Won't save me much money though, I think I'll need a new graphics card to run it properly, it's maxing my 2Gb one out.

[reh321]

I've been using some of Adobe's AI processing filters for a while. Mostly they are very good, sometimes they fail spectacularly. One of the things that seems to work quite well is what they call Enhancement, which doubles the pixel count, taking, for example, a 6mp *istD image to 24MP, ot a K1 image to 144 MP.

The technology is imperfect, but it gets much better with every new version. The failures are fewer and farther between.

Virtually always I reduce the size {number of pixels} of an image.

Once I did ‘upscale’ an image, but then I carefully looked over it to make sure it didn’t have anything unexpected.

[Dartmoor Dave]

Late to this one, not unusual. I've been trying out the trial. I've often thought that some of the results from the Ai software doesn't look natural and that's still true. My results are a mixed bag but I've had enough good results now to make the hop and buy it. In particular I've been trying it on scanned slides. Some it's failed miserably at, really enhancing the noise detail instead of removing it, others it's restored beautifully. One slide suddenly looked like it did on the slide. It's a slide I've scanned numerous times on a film scanner and by camera. A really high dynamic range and nothing has managed to get it right. I ran it through Gigapixel AI and voila, suddenly I was looking at the slide as I was seeing it held up to the light with a magnifier on it. I couldn't quite believe it. Detail, colours, dynamic range, all spot on. I did a few, what the.... double takes. It's also working pretty well on some older Canon S30 3Mp files.

I shall await one of their offers and purchase it. Won't save me much money though, I think I'll need a new graphics card to run it properly, it's maxing my 2Gb one out.

I agree completely. Gigapixel AI is likely to produce false textures and unnatural looking results with files from more recent digital cameras over about 12 megapixels, but give it a jpeg from an ancient 3 or 4 megapixel sensor and the results are truly amazing -- huge improvements in detail and sharpness that usually look completely natural.

And as you say, it can work wonders with film, giving 35mm originals an almost medium format look. Film purists are likely to howl at the idea, but personally I'm running my 35mm negatives through Gigapixel routinely nowadays. I haven't actually tried it on any of my old Kodachromes yet, so I think I'll have to give that a go too.

[3by2]

I agree completely. Gigapixel AI is likely to produce false textures and unnatural looking results with files from more recent digital cameras over about 12 megapixels, but give it a jpeg from an ancient 3 or 4 megapixel sensor and the results are truly amazing -- huge improvements in detail and sharpness that usually look completely natural.

And as you say, it can work wonders with film, giving 35mm originals an almost medium format look. Film purists are likely to howl at the idea, but personally I'm running my 35mm negatives through Gigapixel routinely nowadays. I haven't actually tried it on any of my old Kodachromes yet, so I think I'll have to give that a go too.

Well it was a Kodachrome that failed for me but I wouldn't read anything into that, it's in a fairly bad state. Fujichrome has worked well, though I've had some strange colour shifts. I've taken to adjusting colour balance and levels and exporting to a TiFF to let Gigapixel work on that. Too early yet for definitive workflow conclusions though, it takes such a long time on my PC that I'm not getting through many, loads of Ektachromes, C41 negs and B&W negs still to look at yet. It's definitely worth trying . I'd kinda dismissed it from some of the odd results I'd seen but then I stumbled on a video on youtube about printing from older lower resolution sensors and it was a lightbulb moment. This bloke was working with layers as well to paint in the best from the Gigapixel image whilst leaving the rest original.

[3by2]

Here we go, it does work on Kodachromes, a screenshot at 100%. This is from 1986, scanned on several things but this is the K5 camera copy. The top bit of the Gigapixel image has got the trial logo infringing. It's qute a remarkable difference. I'd buy it just for this image, which I took from the crows nest of a sailing schooner I was on. This is just a straight 1:1 conversion, no upscaling or downscaling.