Created a 'Pixel and PPI Print Chart'

[Immaculens]

Have been learning and creating this PPI Pixel Print Chart with ChatGPT for a few days. Have been saving them in pdf format.

It started out by me having the issue of losing original images 7 yrs ago due to a ransomware attack and finding some smaller Lrc exported versions of the originals - then upscaling them in Photoshop. I then wondered how big I could print these upscaled images and then learned about PPI - and so we made this chart for my reference.

If anyone knows any better - I would like to know if this chart seems reasonable. If its good, feel free to copy it or I can send you the pdf. The notes are pretty basic, just for simple reference.

If there is a more appropriate location for this, feel free to move it.

 

[paddler4]

I'm a little confused by this.

The table is just the number of pixels divided by ppi = inches. The real issue is what's at the bottom: how big can you safely print? The two parts at the bottom contradict each other. The first says that the number in the table are "approximate maximum print sizes." The rule at the bottom says "doubling pixels is safe". Did Chat GPT come up with this? The first is clearly wrong, and IMHO, the second is often too conservative.

There is no one answer because it depends on viewing distance, the nature of the image, etc. However, even before things like Gigapixel and Super Resolution, good software did a pretty good job of upscaling, and IMHO, you could often go considerably bigger than doubling the dimensions implied by the native resolution of the printer. Smugmug's guidelines are here: https://www.smugmughelp.com/hc/en-us/articles/18212578443924-Resolution-requirements-for-printing. I do my own printing, and I've gone well beyond a multiple of 2. For example, the image below I captured with a little 12 MP MFT LX-100 that I sometimes carry when I'm traveling and don't want to lug my gear. The final version is 2056 pixels on the long dimension. I print with a Prograf Pro 1000, which has a native resolution of 300 ppi. So, the chart would say 6.85 inches, and doubling that would be 13.7 inches on the long side. I printed it 19 inches on the long side, a multiple of almost 2.8, and exhibited it in a gallery. This was years before things like Super Resolution, and in fact, I printed it from Lightroom, letting Lightroom use its own upscaling algorithm. If you closely examine it next to similar prints from higher-pixel-count files, you can see that it's not quite as detailed, but the number of people who have commented negatively in the year I exhibited it and the many years it has hung on my wall is zero. The only comments I've received about the print have been positive. One customer at the gallery commented that it was his favorite of the prints I had on exhibit.

On the other hand, I do a lot of macro work where extremely fine details can matter. With those, I try for a small multiple.

With AI-based upscaling, one might be able to go appreciably higher, but as with anything based on AI, it's a role of the dice. If the training set is good relative to the image you are working on, you may get good results. I've also gotten some results that are so bad I discarded the upscaled image and went back to the original.

It also depends on the intended use and audience. A few years ago, I was sent an awful, low-res shot of a woman finishing a race. Her father was very proud, as she had been badly overweight for most of her life, and this race culminated a couple of years of disciplined effort to lose weight. He asked if I could print an 8 x 10. I did what I could, but the original was really small. I no longer have it, so I can't say precisely. In the end, I got a print that I considered passable but clearly not up to my standards. He was absolutely delighted and immediately framed it and hung it on his wall.

So my guideline would be a little more flexible. If you are going much past 2, you may get problems, but it varies. And in my experience, printing from Lightroom, which does all of the upscaling for you, yields excellent results, not apprecialy different from what I've gotten with the much more complex Photoshop routine.

 

[Bobtographer]

Years ago, I e-mailed this photo of my daughter to my dad. He printed it out at 9x6 for my mum. Until you get in close, you can't tell that it's low res- 800px longest edge..

 

[Immaculens]

I'm a little confused by this.

The table is just the number of pixels divided by ppi = inches. The real issue is what's at the bottom: how big can you safely print? The two parts at the bottom contradict each other. The first says that the number in the table are "approximate maximum print sizes." The rule at the bottom says "doubling pixels is safe". Did Chat GPT come up with this? The first is clearly wrong, and IMHO, the second is often too conservative.

There is no one answer because it depends on viewing distance, the nature of the image, etc. However, even before things like Gigapixel and Super Resolution, good software did a pretty good job of upscaling, and IMHO, you could often go considerably bigger than doubling the dimensions implied by the native resolution of the printer. Smugmug's guidelines are here: https://www.smugmughelp.com/hc/en-us/articles/18212578443924-Resolution-requirements-for-printing. I do my own printing, and I've gone well beyond a multiple of 2. For example, the image below I captured with a little 12 MP MFT LX-100 that I sometimes carry when I'm traveling and don't want to lug my gear. The final version is 2056 pixels on the long dimension. I print with a Prograf Pro 1000, which has a native resolution of 300 ppi. So, the chart would say 6.85 inches, and doubling that would be 13.7 inches on the long side. I printed it 19 inches on the long side, a multiple of almost 2.8, and exhibited it in a gallery. This was years before things like Super Resolution, and in fact, I printed it from Lightroom, letting Lightroom use its own upscaling algorithm. If you closely examine it next to similar prints from higher-pixel-count files, you can see that it's not quite as detailed, but the number of people who have commented negatively in the year I exhibited it and the many years it has hung on my wall is zero. The only comments I've received about the print have been positive. One customer at the gallery commented that it was his favorite of the prints I had on exhibit.

On the other hand, I do a lot of macro work where extremely fine details can matter. With those, I try for a small multiple.

With AI-based upscaling, one might be able to go appreciably higher, but as with anything based on AI, it's a role of the dice. If the training set is good relative to the image you are working on, you may get good results. I've also gotten some results that are so bad I discarded the upscaled image and went back to the original.

It also depends on the intended use and audience. A few years ago, I was sent an awful, low-res shot of a woman finishing a race. Her father was very proud, as she had been badly overweight for most of her life, and this race culminated a couple of years of disciplined effort to lose weight. He asked if I could print an 8 x 10. I did what I could, but the original was really small. I no longer have it, so I can't say precisely. In the end, I got a print that I considered passable but clearly not up to my standards. He was absolutely delighted and immediately framed it and hung it on his wall.

So my guideline would be a little more flexible. If you are going much past 2, you may get problems, but it varies. And in my experience, printing from Lightroom, which does all of the upscaling for you, yields excellent results, not apprecialy different from what I've gotten with the much more complex Photoshop routine.

Thank-you so much for your educated reply paddler4. My own experience and supposed knowledge regarding PPI was quite flawed, I realized quickly, in chats with chatgpt. The chart was the result of about 3 days of me gleaning any truth nuggets I could, but in the end the chart result was largely influenced by my actual questions and understanding.

As I mentioned, my quest began due to lost original images and how I could perhaps rescue them with typically little 100kb, 900 long edge pixels, 72ppi web images. In Ps I ended up usually doubling the size to 1800 pixels at 300ppi (300ppi until I just days ago realized that 300ppi was limiting my print size to perhaps 4x6, if I understand correctly). Chat GPT suggested 240ppi so I could print larger as long as it was expected the viewer would not be viewing close but rather at a bit of distance.

I thought your reply was so nutrient dense that I wondered how ChatGPT would evaluate and respond to your reply, so I plugged in your reply and thought it would add to the discussion here. I took screen shots of my original question and the reply from ChatGPT. See images below:














See next post for part 4 of ChatGPT reply:

 

[Immaculens]

That concludes the full reply from ChatGPT. This chart, notes, and reply to your post are all created by the 'Free' version of ChatGPT. I mention this because the plus (or paid) version would no doubt be more thorough in details.

I still want to create a reference chart, so I welcome any further discussion.

 

[paddler4]

Chat GPT is now closer, but it still isn't quite right. One of its suggested revisions is "with high-quality resampling,(Lightroom, Photoshop, AI tools), prints up to 2 x the listed sizes [emphasis in the original] are often visually excellent." True, but as my example shows, and as the Smugmug guidelines say, even without AI tools, prints above 2x the listed sizes are often excellent.

I'm no AI experts, but I think the problem with AI is that there is no real "I". It's based on scraping huge volumes of text, some of which are wrong or misleading, and it's all just a mass of probabilities. There is no expert hidden there evaluating the stuff that's scraped. Often (increasingly often) it will provide excellent answers. Sometimes it doesn't, and sometimes it just makes stuff up ("hallucinates"). So IMHO, it's often a great place to start, but not always a great place to end.
,
Some time ago, a friend asked for a PDF of one of my books and fed it to Google's NotebookLM to create a podcast. It was for the most part unnervingly good, but it got a couple of central points dead wrong.

 

[Immaculens]

Chat GPT is now closer, but it still isn't quite right. One of its suggested revisions is "with high-quality resampling,(Lightroom, Photoshop, AI tools), prints up to 2 x the listed sizes [emphasis in the original] are often visually excellent." True, but as my example shows, and as the Smugmug guidelines say, even without AI tools, prints above 2x the listed sizes are often excellent.

I'm no AI experts, but I think the problem with AI is that there is no real "I". It's based on scraping huge volumes of text, some of which are wrong or misleading, and it's all just a mass of probabilities. There is no expert hidden there evaluating the stuff that's scraped. Often (increasingly often) it will provide excellent answers. Sometimes it doesn't, and sometimes it just makes stuff up ("hallucinates"). So IMHO, it's often a great place to start, but not always a great place to end.
,
Some time ago, a friend asked for a PDF of one of my books and fed it to Google's NotebookLM to create a podcast. It was for the most part unnervingly good, but it got a couple of central points dead wrong.

When you have a moment - please give me your thoughts on this version. I did a version 13. Chat GTP is friggin dozy today, holy cow...

I have not, admittedly, read the notes for a while. I suspect they are overly simplified lol...

 

[paddler4]

Frankly, I think this isn't a productive way for you to go about it. If you really want to delve into the details of printing the best possible way at different sizes, read what one of the real experts says, not what ChatGPT comes up with after scraping large amounts of text of varying accuracy. One example is the things written by Jeff Schewe, e.g., http://www.schewephoto.com/workshop/pdfs/GreatOutput-GreatPrints.pdf.

However, for almost all people, Schewe's level of detail is simply unnecessary. I've printed for years, sometimes competitively, and have almost never done more than what's in the next paragraph.

For most printing, just set the print size in Lightroom and set the print resolution to the native resolution of the printer. If the image is really low resolution, you might want to try using an AI-based upscaling tool to see whether it helps before doing this. In almost all cases, at reasonable sizes--say, 5 x 7 to 7 x 22--this will work just fine.

So, what's "really low resolution"? One doesn't need a table, as it's just simple arithmetic. Divide the dimensions of the base image in pixels by the native resolution of the printer. That's the biggest print you can get without upscaling. So, for example, if I have an image that is only 1200 PX across, and my printer has a native resolution of 300 dpi, so I can print 4 inches with no upscaling. If the print you want is much more than 2 or 3 times that big, you may need to do more work. But if the print is not going to be examined up close, it may be just fine.

If you want to go beyond that, it gets more complicated. There is both a printer resolution and a print resolution. Printers don't give you lots of options. For example, my Canon printer, which is a professional-level printer, allows two choices of printer resolution, 300 (the default) and 600 dpi. As Schewe wrote--I tested this after reading his article--the higher printer resolution produces somewhat sharper prints if you are downsizing. Not relevant to your chart, which is about upscaling.

For print resolutions, Schewe also argues that one has limited choices, although more:

For Ink jet output, make sure the file resolution is
set to a number equally divisible into the printer resolution.

But again, this is all unnecessary for almost all printing. I've printed for years, sometimes competitively, and I have never had anything but positive comments about the quality of prints I've produced this way. Many professionals still do the complicated Photoshop procedure, but when I did an A/B test, it didn't provide appreciably better results than simply using Lightroom's built-in upscaling algorithm. However, your mileage may vary.

Frankly, for people looking seriously at a print, lots of other things matter much more--e.g., good tonal variation, good colors, an appropriate choice of paper, yada yada.

 

[Immaculens]

Frankly, I think this isn't a productive way for you to go about it. If you really want to delve into the details of printing the best possible way at different sizes, read what one of the real experts says, not what ChatGPT comes up with after scraping large amounts of text of varying accuracy. One example is the things written by Jeff Schewe, e.g., http://www.schewephoto.com/workshop/pdfs/GreatOutput-GreatPrints.pdf.

However, for almost all people, Schewe's level of detail is simply unnecessary. I've printed for years, sometimes competitively, and have almost never done more than what's in the next paragraph.

For most printing, just set the print size in Lightroom and set the print resolution to the native resolution of the printer. If the image is really low resolution, you might want to try using an AI-based upscaling tool to see whether it helps before doing this. In almost all cases, at reasonable sizes--say, 5 x 7 to 7 x 22--this will work just fine.

So, what's "really low resolution"? One doesn't need a table, as it's just simple arithmetic. Divide the dimensions of the base image in pixels by the native resolution of the printer. That's the biggest print you can get without upscaling. So, for example, if I have an image that is only 1200 PX across, and my printer has a native resolution of 300 dpi, so I can print 4 inches with no upscaling. If the print you want is much more than 2 or 3 times that big, you may need to do more work. But if the print is not going to be examined up close, it may be just fine.

If you want to go beyond that, it gets more complicated. There is both a printer resolution and a print resolution. Printers don't give you lots of options. For example, my Canon printer, which is a professional-level printer, allows two choices of printer resolution, 300 (the default) and 600 dpi. As Schewe wrote--I tested this after reading his article--the higher printer resolution produces somewhat sharper prints if you are downsizing. Not relevant to your chart, which is about upscaling.

For print resolutions, Schewe also argues that one has limited choices, although more:



But again, this is all unnecessary for almost all printing. I've printed for years, sometimes competitively, and I have never had anything but positive comments about the quality of prints I've produced this way. Many professionals still do the complicated Photoshop procedure, but when I did an A/B test, it didn't provide appreciably better results than simply using Lightroom's built-in upscaling algorithm. However, your mileage may vary.

Frankly, for people looking seriously at a print, lots of other things matter much more--e.g., good tonal variation, good colors, an appropriate choice of paper, yada yada.

Thank you so much for all your detail and offering the benefit of your experience! Much appreciated!

 

[paddler4]

Good luck. Learning to print is complicated enough without all the extra stuff.