Wanted - A Few Guinea Pigs for Testing

[Fred Weiss]

In the last couple of months, I've been acquiring and experimenting with various image editing software that features the use of artificial intelligence. While some of it appears to me to be overrated, a great deal of it may well be revolutionary, particularly in terms of saving a lot of time while delivering professional results.

What I would like to determine is whether or not some of these new capabilities would be viable to offer as an image/photo enhancement and modification service that would be of interest to wide format printers. Kind of like what Eric, The Vector Doctor does for vector images except this would be for photos and raster images.

The various services offered would include:

• Lost detail recovery
• Noise and artifact reduction
• Focus, exposure and sharpness improvement
• Selective lighting, detail and color enhancement
• Unwanted object removal from backgrounds
• Background removal
• Sky replacement, background replacement and general compositing
• Portrait retouching
• Resizing up to 1,000%

So if you have a photo or image that might benefit from the above that you would like to actually print for a client, send it to me attached to an email along with who you are and a description of what you would like done to the image. Please ... Original, unmodified images or photos ONLY with a minimum size of 800 pixels for the longer dimension.

There will be no charge for my services at this time but it will be limited to the first 10 requests received. If your image is accepted, I do ask that you reply back to this thread with your unfiltered opinions of the value of what you receive back.

Send emails to: imageservices@allcompu.com

Discussion of artificial intelligence and the proposed image enhancement service are invited and welcome in this thread.

 

[bannertime]

I made a post about an online tool when it came out, but no one wanted to talk about it. I've been using Preserve Details 2.0, for scaling, since it's release in Photoshop. Hadn't needing to try anything more advanced. End up using another image or get access to a higher quality original.

https://signs101.com/threads/lets-enhance-using-neural-networks-to-improve-image-quality.143567/

 

[Fred Weiss]

When it comes to digital enlargements, that is the final step in my workflow. For it, I use either Adobe Photoshop Preserve Details 2.0, OnOne Resize or Topaz Gigapixel AI. Each goes about what it does in its own way and each comes out with differing results. none of them is an every time choice over the other two. So depending on the image and what is being looked for in the enlargement, will determine which one will end up being used.

In the case of JPG originals, my workflow starts with Topaz JPG to Raw. This gives me a much better starting point because most artifacts are removed and much detail is recovered that was lost as soon JPG compression was introduced. The saved Raw file it produces is typically 30 to 35 times larger in size (not dimensions) than the JPG it was made from.

From there we go through steps to reduce noise, improve sharpness and focus, and adjustment of light, color and details. Once there we move into AI enhanced adjustments before ever getting to enlargement.

In working with a number of AI products, I came to realize that there are several very different benefits to be had. Gigapixel AI got a lot of hype over its ability to "imagine" what detail needed to restored. The result was sometimes great and sometimes a total wrong guess. To me, the greatest benefit of AI is its ability to recognize selective elements in an image. That may be a model's eyes or lips, a sky, vegetation in a landscape, etc. so that I don't have to select or mask to make a model's lips redder or change out an entire sky for another. AI also decides what details to enhance or blur to reduce.

It might be interesting to try what you tried with the automated website in my workflow to see if a human controlled, semi-automated workflow does a better job.

Here's a before and after I did today:

 

[jfiscus]

I love the Topaz products, put them to use at least every week now. Incredible results from mediocre input images.

 

[Billct2]

If the service was as reasonably priced as the vector doctor I would probably use it.

 

[Fred Weiss]

If the service was as reasonably priced as the vector doctor I would probably use it.

Not locked in but I've got $19.95 tentatively penciled in for the whole enhancement workflow with resizing and a couple of changes allowed. Individual items in the $5 to $10 range.

Reasonable enough?

 

[Fred Weiss]

Would you like to take a stab at some oversized eps's that my customer provided? Whoever had drawn up a mural I installed took some 300dpi b/w photograph scans and blew them up to 9' tall, then increased the dpi to 300 again. One panel ended up being 2gb and pixels the size of my thumb(exaggeration), so it may be too little there to actually improve. Let me know if it falls in the same category and I'll get the art uploaded if so.

It doesn't sound like something that fits into what I'm offering. We're looking to work with unmodified originals and deliver enhanced and enlarged versions back to you.

 

[unclebun]

Fred, on your example, I think something's overdone in the woman's skin tones. The center of her face looks sunburned pink.

 

[Fred Weiss]

Fred, on your example, I think something's overdone in the woman's skin tones. The center of her face looks sunburned pink.

I kind of liked the look. Here's an adjusted mod ... do you like it better?

 

[Gino]

I wasn't going to say anything, but since the coloring was brought up..... the process you are using seems to age the person. She seems to have more wrinkles, deeper creases and her hair appears more greasy. Is that due to lack of control or not noticing ?? You're using a flesh tones and that's where these things tend to show up the most. The coloring is much improved, also. Before, you had too much color on for life-like appeal. Keeping the background colors separate from the foreground colors really helps.

This whole thing reminds me of how television uses cameras for close-ups and they can use the same camera and fuzzy out people's wrinkles and laugh-lines by changing the lens a little bit.

 

[Fred Weiss]

I wasn't going to say anything, but since the coloring was brought up..... the process you are using seems to age the person. She seems to have more wrinkles, deeper creases and her hair appears more greasy. Is that due to lack of control or not noticing ?? You're using a flesh tones and that's where these things tend to show up the most. The coloring is much improved, also. Before, you had too much color on for life-like appeal. Keeping the background colors separate from the foreground colors really helps.

This whole thing reminds me of how television uses cameras for close-ups and they can use the same camera and fuzzy out people's wrinkles and laugh-lines by changing the lens a little bit.

There's a certain amount of "unintended changes" that happen when any adjustment is done. Color levels and hues, in particular, tend to change a bit even when the adjustment has nothing to do with color. I don't think she looks older or her hair looks greasy. What I'm seeing is an increase in lens focus and sharpness with an unintended deepening and enriching of color. Some AI portrait features were also used to separately enhance her eyes, the whites of her eyes, her eyebrows, her lips and the width of her face was slightly narrowed to offset widening of her face due to smiling. If you look at the edges of her hair, you will get an idea of how much the focus has been improved as compared to the original.

One of my main concerns in exploring the viability of this as a service is communication between the client and myself. For example, are the enhancements to be subtle, middle of the road or extensive?

We recently replaced our high definition living room TV with a UHD model. The first thing we noticed was the sort of things you point out. Deeper creases etc. and more crispness were visible because we we looking at higher resolution. Whether or not such changes are desirable in any given job is highly subjective and problematic to handle.

 

[unclebun]

Yeah, it's less pronounced the second time, but it's still doing it. It's like there's a stripe of brighter pink/magenta down the center of her face. I think I can see in the original what is getting enhanced. Many women (and men too but we don't care about the color of our faces) have more redness in the center triangle of the face. It comes from more blood vessels. But many/most women spend time and money with their makeup to even out the skin tone so you can't notice it so much.

The way the skin tone on her chest came out is fine. Also, her blouse no longer looks like one color overall. The sleeve in front is almost blue, and the one in back is more like the original brownish tone. It's probably the same thing being done there as on the face. It's seeing some difference in tone due to lighting and accentuating it too much.

The sharpening/focus treatment is fine, but I think I see what Gino is referring to. Over her left shoulder, the enhancement of her hair has created such contrast lines that it makes the hair look like it would if it were a little wet, or greasy. Like the hair is clumping up.

Anyway, the tool certainly seems powerful and does enhance sharpness of focus. The kinds of things we're talking about I'm sure can be adjusted, and personal taste/preference will play into that.

 

[Andy_warp]

I think there would be a great market for this in the grand format sector...we get notoriously bad images that started as a low res jpeg and ALL of the artifacts get upsampled and enlarged with the "good" stuff.
As Bannertime stated the preserve details 2.0 is incredible...but you can only do so much with something that's already been compressed.

If you go right from a raw file, it is astounding.

...that said, I see the trends leaning more towards softer images these days. I like sharp and crisp, but not sure I've ever been in style...

 

[Fred Weiss]

I think there would be a great market for this in the grand format sector...we get notoriously bad images that started as a low res jpeg and ALL of the artifacts get upsampled and enlarged with the "good" stuff.
As Bannertime stated the preserve details 2.0 is incredible...but you can only do so much with something that's already been compressed.

If you go right from a raw file, it is astounding.

...that said, I see the trends leaning more towards softer images these days. I like sharp and crisp, but not sure I've ever been in style...

It would be interesting to run one or two of your low res jpegs through Topaz JPG to Raw and then have you work it through the rest of the way to see if it helps. Send me a couple of images and I'll process them and return them.

 

[Ldireprophil]

I sent Fred a random (unedited pic) that a customer had given me the other day. They actually texted me the jpg and I didn't even look at it. I have Adobe CC 2020, GIMP 2, Inkscape, Fotophire, etc., but as a small shop don't have the time to "enhance" each and every photo/image I receive. This customer wants this pic taken in Honduras printed on canvas and wrapped, no problem. I would have to say that for the super quick reply I got from Fred with the enhanced image would certainly be worth the few $'s that he is thinking of charging. It would free me up to do other things and I'm sure I would send him many just like this for the same reason.

 

[myront]

Not bad. I often do the same with Photopaint and fairly quickly.
The highlight in the hair, especially on top, have too much of a blue hue

 

[Pauly]

When it comes to digital enlargements, that is the final step in my workflow. For it, I use either Adobe Photoshop Preserve Details 2.0, OnOne Resize or Topaz Gigapixel AI. Each goes about what it does in its own way and each comes out with differing results. none of them is an every time choice over the other two. So depending on the image and what is being looked for in the enlargement, will determine which one will end up being used.

In the case of JPG originals, my workflow starts with Topaz JPG to Raw. This gives me a much better starting point because most artifacts are removed and much detail is recovered that was lost as soon JPG compression was introduced. The saved Raw file it produces is typically 30 to 35 times larger in size (not dimensions) than the JPG it was made from.

From there we go through steps to reduce noise, improve sharpness and focus, and adjustment of light, color and details. Once there we move into AI enhanced adjustments before ever getting to enlargement.

In working with a number of AI products, I came to realize that there are several very different benefits to be had. Gigapixel AI got a lot of hype over its ability to "imagine" what detail needed to restored. The result was sometimes great and sometimes a total wrong guess. To me, the greatest benefit of AI is its ability to recognize selective elements in an image. That may be a model's eyes or lips, a sky, vegetation in a landscape, etc. so that I don't have to select or mask to make a model's lips redder or change out an entire sky for another. AI also decides what details to enhance or blur to reduce.

It might be interesting to try what you tried with the automated website in my workflow to see if a human controlled, semi-automated workflow does a better job.

Here's a before and after I did today:
View attachment 144362

I just want to add my 2c in.
As someone who does a lot of printing for photographers, and also a photographer myself, If i had to use software to blow up my low res image or get someone to do it for me. I'd expect the image to look the same with the exception of more detail. changing colour tones wouldn't be acceptable as it's not what the photographer intended the photo to look.

In the image you posted, the original appears to have a nice soft golden hour/warm tone to it, slightly muted blacks to get that smooth look. Although the detail is nicer and acceptable in the modified version, The colours are not. There's a lot of unnatural colour in her face such as blues in the highlights when there's a warm skin tone. There's also green in her hair, The grass is far to green (on the cool/blue side) and so on.

I use Topaz gigapixel & Adobe Photoshop Preserve Details 2.0 when i have to enlarge images.
depends on the image im working with i'll choose one or the other. sometimes a bit of both. & i'll never change the colours of the original image as it's not how it was intended to be.

 

[Fred Weiss]

I just want to add my 2c in.
As someone who does a lot of printing for photographers, and also a photographer myself, If i had to use software to blow up my low res image or get someone to do it for me. I'd expect the image to look the same with the exception of more detail. changing colour tones wouldn't be acceptable as it's not what the photographer intended the photo to look.

In the image you posted, the original appears to have a nice soft golden hour/warm tone to it, slightly muted blacks to get that smooth look. Although the detail is nicer and acceptable in the modified version, The colours are not. There's a lot of unnatural colour in her face such as blues in the highlights when there's a warm skin tone. There's also green in her hair, The grass is far to green (on the cool/blue side) and so on.

I use Topaz gigapixel & Adobe Photoshop Preserve Details 2.0 when i have to enlarge images.
depends on the image im working with i'll choose one or the other. sometimes a bit of both. & i'll never change the colours of the original image as it's not how it was intended to be.

I hear where you're coming from and would agree that such restrictions are appropriate in some cases. I cannot agree that it should be a hard and fast rule though. Not because I think it's great to change any more than necessary but because it is often a better looking result when things like color get modified. And it's often the case that there is no particular intent on the part of the taker of the photo other than to capture an image of the content of the shot.

In the case of the woman posing, I felt like I was not dealing with an original as shot ... rather a photo that had been softened to chase a mood and dulled down with a golden-tan color cast. It was a late night attempt to lose that and make it more realistic and attention grabbing. It was posted prematurely and it should have been worked with a while longer. The second photo, however, of the man waving from a bench in Post #16 is much more of what I have to offer. It was shot from a smart phone and measured about 800 pixels on its long side. In it, the sun is somewhere behind him and his features are almost lost in the silhouetting that resulted. The After version looks quite different in terms of color, detail, exposure, focus and sharpness. Whether those differences are better or not is for each individual to decide. It is a better example of what one can expect from my efforts. The following images are a before and after with the after sized down to the size of the original. The second is a link to the after image at 5,000 pixels in width ... an enlargement of about 625%.

Link to Full Size Image

I'm sure there are still improvements to be had but I think it succeeds in making an unusable snapshot into a photo that will print well and please the client.

 

[ddarlak]

i wouldn't need this service very often, but i have had a few cases like the Honduras guy in the past and if the charge was $19.95 i would gladly pay for it.

 

[Pauly]

I hear where you're coming from and would agree that such restrictions are appropriate in some cases. I cannot agree that it should be a hard and fast rule though. Not because I think it's great to change any more than necessary but because it is often a better looking result when things like color get modified. And it's often the case that there is no particular intent on the part of the taker of the photo other than to capture an image of the content of the shot.

In the case of the woman posing, I felt like I was not dealing with an original as shot ... rather a photo that had been softened to chase a mood and dulled down with a golden-tan color cast. It was a late night attempt to lose that and make it more realistic and attention grabbing. It was posted prematurely and it should have been worked with a while longer. The second photo, however, of the man waving from a bench in Post #16 is much more of what I have to offer. It was shot from a smart phone and measured about 800 pixels on its long side. In it, the sun is somewhere behind him and his features are almost lost in the silhouetting that resulted. The After version looks quite different in terms of color, detail, exposure, focus and sharpness. Whether those differences are better or not is for each individual to decide. It is a better example of what one can expect from my efforts. The following images are a before and after with the after sized down to the size of the original. The second is a link to the after image at 5,000 pixels in width ... an enlargement of about 625%.

View attachment 144501
Link to Full Size Image

I'm sure there are still improvements to be had but I think it succeeds in making an unusable snapshot into a photo that will print well and please the client.

Sorry missed the link in that post, but that example is much better. I know it's also difficult to enlarge images of people with out them looking like a cartoon.
I can see what you're trying to do and it's great.
But i'll stand by where colour grading should be left alone unless specified by the client. If you're offering this as a service, id suggest that you provide an option to "enhance colour"