How to get a high resolution image when the master is on a CD

A CD master is often already the high-resolution file. Upscale 4× only when the long edge is actually too short for the print.

My sister wanted a 12 inch print from her anniversary site and asked how to get a high resolution image. The site has served the same pictures since 2011. Right-click, save, and the JPEG that lands is 1200 by 800 pixels, about 210 KB. She assumed the site had the good copy because the site is where the family looks. The good copies are on a CD in our father's desk, in a paper sleeve, marked with a marker "CAM 2009." The disc is dull and a little scratched. It still reads. That disc is the answer, and the 1200 pixel download is a thumbnail that learned to dress like a webpage.

I saved one site image and opened the info window before I drove to the desk. 1200 by 800. For a 12 inch wide print at 300 pixels per inch I want about 3600 pixels on the long side. She has a third of that. At 4 inches wide, 1200 divided by 4 is 300, and the picture would be fine. She does not want 4 inches. She wants the group at the table, in a frame she already bought, 12 by 8. The website cannot supply that frame. A model can guess it. The camera already recorded it, if we can find the card dump.

The uploads folder, which she still has on an old host, confirms the loss. The largest derivative is 1200 by 800. There is a 1024 version, a 768, a 300. There is no 3872. The theme had a maximum width and it obeyed it on the way in, then discarded the larger file. How to get a high resolution image, here, is a search of a drawer, not a slider. I say that before anyone pays for an upscale of a picture we can copy.

TL;DR. The anniversary site only serves 1200 by 800 derivatives, about 210 KB, because the theme scaled uploads and dropped the originals. The camera files are on a CD labeled CAM 2009, at 3872 by 2592, about 3.8 MB each. Thirty-six of the forty site photos match those files by EXIF time. A 12 inch print at 300 pixels per inch wants about 3600 pixels, which the disc has and the site does not. Two files on the disc are unreadable, and one guestbook page was never on the disc. Those are the only frames where an upscale is even a discussion. Finding the master beats inventing one.

What 1200 pixels can honestly print

I divide before I editorialize. 1200 pixels divided by 4 inches is 300 pixels per inch. Divided by 6 inches it is 200. Divided by 8 it is 150. Divided by 12 it is 100. At 100 pixels per inch, a face in that group photo is soft in the hand and acceptable only if the viewer stands back and expects a poster, not a record. She will hang the frame in a hall where people stop at about 3 feet. At 3 feet I want the 300, or something close. One hundred will look like the website, which is what she is already unhappy about.

The 210 KB size is the second hint. A 3872 by 2592 JPEG from that camera, at the quality it used, sits around 3.8 MB in this set. A file ten to twenty times smaller, at a fraction of the pixel count, has been resampled and compressed again. It can still look neat on a phone. Phones are small. The frame is not a phone. I had her hold a 4 by 6 test of the website file next to the empty 12 inch frame. The test looked all right. The empty frame was the argument. We did not fill the large frame with a stretched 1200.

Page source was worth ten minutes even though the disc was the real hope. Sometimes a builder leaves the original under a different path. Hers did not. The img element and the srcset pointed at the same family of derivatives, and the widest candidate was 1200. I have seen a 2400 hiding in a srcset while the visible image was small. Not this host. I wrote "largest on server 1200" in the notes so I would not browse the media library a second time out of optimism.

The disc, the folder, and the match

The sleeve was under a stack of manuals. The disc label is marker, not a print. I copied the whole disc to a drive before I opened pictures, so a struggling drive would not be asked to retry all afternoon. The copy took a while. Two files threw read errors and I logged their names. Everything else came across. The photos live in a DCIM folder, then a subfolder of numbered JPEGs with an IMG prefix and a number. There are 214 JPEGs on the disc. The site used about 40. The rest are outtakes she was right to skip, and they are still better records than the derivatives of the frames she kept.

A typical keeper is 3872 by 2592, 3.8 MB, shot on a 2009 afternoon, ISO 200, one two-hundredth, f/4 or so, embedded color from the camera. 3872 divided by 12 inches is about 323 pixels per inch. The long side of her frame is covered with room to crop. 2592 divided by 8 is 324. The short side is the limit, and the limit is still above 300. The disc file is not "infinite resolution." It is enough for this frame, from life, with the lace and the glasses and the cake lettering that the 1200 file turns into bumps.

The site had renamed files to anniversary-01 and up, stripped of the camera numbers. I matched on the EXIF time. For 36 files the time on the derivative, which the theme had kept, matched a disc file within the same second. One pair of shots was taken a second apart, almost the same table, and the times could have been assigned either way if I had been sloppy. In one of them a guest wears a red hat and turns toward the cake. In the other the hat is on a chair. The website caption talks about the hat on his head. I assigned the caption to the file where the hat is on his head. A small thing, and it is the difference between a correct print and a confident swap.

Four site images did not match any disc time. Two are guestbook pages, photographed or scanned later, not part of CAM 2009. Two are from a phone of that era, already 1600 pixels on the long side even in the "original" she found in an old inbox, and the site had scaled those down to 1200 as well. The inbox file, 1600, is the master for those two. I did not upscale the site's 1200 when a 1600 existed in mail. The gain is small and it is real, and real wins.

Read errors stay read errors

Image 2410 and image 2411 would not copy. I cleaned the disc surface with a dry cloth, tried another drive, and got the same errors. The site still has those two frames at 1200 by 800. They are a speech and a toast, slightly soft even in memory because the room was dim. I printed each at 4 by 2.7 inches from the 1200 file, which spends the pixels at about 300 per inch, and I stopped. She asked about a 2 times enlargement for an 8 inch width. I ran that test on the 4x upscale path's smaller sibling, a 2 times pass, and the microphone in the speech frame grew a grille pattern the real mic, still in a closet, does not have. I showed her the closet mic and the invented grille. We kept the 4 inch prints. The scratches on the disc are allowed to have won those two frames.

The guestbook, which was paper, not a camera file

The guestbook page on the site is 1200 by 800 and the handwriting is already broken. Upscaling handwriting is how you get letters that look confident and say nothing. I asked if the book was still in the house. It was, in a bin with the centerpieces. The page is about 8 inches wide. I scanned it at an optical 600, and 8 times 600 is 4800 pixels. The scan shows the pen pressure. The website crop had thrown that away. We printed the scan. How to get a high resolution image, for this page, was a flatbed, not a model.

One other guestbook spread had been tossed with the napkins, she thinks. Only the 1200 file remains. I did not invent the missing signatures. I printed that spread small, at the 4 inch standard, and I left a note in the album that the book is gone. A careful 2 times pass made the loops of ink smoother and did not make an illegible name legible. Smoother illegibility is not a reading. I discarded the pass. The AI image upscaler is the tool I would use if a modest enlargement of a photographic frame were truly the only path. A signature is not that frame. I would rather have a small true scribble than a large fake letter.

The bin also had a second CD, unlabeled, which turned out to be the caterer's video menus and not photos. I mention it because "a disc in the drawer" is not automatically the camera. I opened directories and looked at pixel sizes. Video menus at 720 pixels wide were a false win for about five minutes. Pixel size ended the celebration. 720 is worse than the website. 3872 is the camera. The folder name DCIM was the other clue, and the EXIF times were the proof.

A table of where each master actually was

Forty website images felt like one problem. They were four problems with different masters.

Group	How many	Master we found	Pixels	Pro I will stand behind
Camera frames on the CD	36	numbered camera files	3872 by 2592	12 by 8 at about 320 ppi
Disc read errors	2	Site derivative only	1200 by 800	About 4 inches wide
Old phone frames	2	Inbox JPEG	1600 on the long side	About 5 inches wide
Guestbook, book survived	1	New scan at 600 ppi	4800 wide	The scan, not the site
Guestbook, book gone	1	Site derivative only	1200 by 800	Small print, no invented letters
Caterer disc	none of the photos	720-wide menus	720	Not a photo master

The first row is the whole victory. Thirty-six frames, no model, no argument about lace. The other rows are where I earn the right to talk about limits. If I upscale the easy 36, I throw away the afternoon I spent at the drawer, and I teach the family that the website was enough.

Before I enlarge anything a website gave me

The order is a scavenger hunt. I like scavenger hunts more than I like repairing a guess.

• Save the displayed image and read its pixels. Do not trust the size it appears on the monitor.
• Check the page source or the media library for a wider derivative, and note the largest number.
• Look for the camera card, the disc, the inbox, and the original paper if the picture is a scan of paper.
• Copy discs before you retouch, and log any file that will not read.
• Match renamed web files by EXIF time, then by a detail you can name, such as a hat.
• Divide the master's long side by the frame's inches. If you are near 300, print and stop.
• Upscale only the frames whose master is gone, and only toward a size you looked at in a proof.
• Do not "recover" text. If a name cannot be read at the true pixel count, it cannot be read.

The hat rule is more general than this party. Pick a detail the caption depends on. If the master and the derivative disagree about that detail, you have the wrong pair or a damaged file. In this set they agreed once I had the hat on the correct second. I did not find a case where the website crop had removed the hat and the disc had it, but I looked, because crops are a kind of loss too. One disc file has more table on the left than the site crop. I printed the wider file. The theme's crop was not sacred. It was a column width.

A drawing with a known pixel count, instead of the wedding party

The group photo has living people in it, and the anniversary is not mine to publish. The stand-in is a generated drawing, 960 by 640 pixels: a round smiling head, a teal rectangular body, a teal panel with a yellow circle, a dark rectangle, three tan bars, and a small green shape on a brown floor. No guest, no hat, no cake.

I use it to talk about "the file you saved" versus "the file that exists." If someone saves this drawing from a page that only offers 320 pixels of it, they do not have the 960. They have a derivative. Going back to the 960 is the repair. Drawing new edges for the yellow circle, from the 320, is a different activity, and it can look clean because the circle is a simple shape. Her cake lettering is not a simple circle. The disc has the lettering. The 1200 file has bumps. I would rather show a public drawing than pretend a model rebuilt the cake.

Nine hundred and sixty pixels divided by 3.2 inches is 300. I can print the drawing about 3.2 inches wide at that standard, and I can print it larger if I accept fewer pixels per inch. The arithmetic is the same arithmetic as the anniversary, with friendlier subject matter. Once the arithmetic is clear, the CD is an obvious place to look, not a nostalgic one.

The 12 by 8, made from the disc, and the prints I refused

The framed picture is the table shot, cropped a little from 3872 by 2592, exported at 3600 by 2400 so the file's pixels match 12 by 8 at 300. I set the tag to 300 because it agrees with the inches. The lace at the hem is a pattern I can count. In the website file the same hem is a gray zigzag. She cried about the lace, briefly, which I did not expect, and then she asked if we could do the rest of the album from the disc. We can. That is 36 frames, a weekend of choosing, not a weekend of upscaling.

I refused a 24 inch wide board of the same table. 3872 divided by 24 is about 161 pixels per inch. For a board seen from 8 feet, 161 can pass. She wanted it behind a table where people stand at 2 feet for photos of their own. At 2 feet, 161 is soft, and a 2 times upscale of even the disc file would be inventing lace at a size that invites inspection. The disc is a high resolution image relative to the website and relative to a 12 inch frame. It is not a billboard. I will say the number 161 out loud rather than let a lab "enhance to fit" the board.

The two unreadable toasts stayed at 4 inches in the album, with a line of type under them saying the camera files could not be read. Type is honest. A sharp fictional microphone is not. The guestbook scan got a full page. The missing guestbook spread stayed small. The two old phone frames got 5 inch prints from the 1600 pixel inbox files. The album is uneven on purpose. Evenness would have required me to damage the good frames down to the bad ones, or to inflate the bad ones up to the good ones. I will not damage the good ones.

Names and the sleeve

The disc went back in the sleeve, and the sleeve went back in the desk, after the copy. The drive folder is anniversary-disc-copy, with the two failed names in a text file. anniversary-site-1200 holds the derivatives I saved for comparison. anniversary-print-12x8 is the framed export. I do not call any upscale "master." The master is the disc copy, and for two files there is no master, which the text file says in a sentence a sibling can read.

I also exported a free plan of the 36 at a small size, made from the disc files, so choices about the album happen without browsing 3872 pixel images on a laptop that chokes. The free plan is a derivative I made on purpose, named as a free plan. The website was a derivative made by a theme that did not ask. The difference is whether anyone kept the original. In 2011 the answer was the CD, luckily. In a lot of families the answer is no, and then we are in the other essay, about living with a small file. This family does not have to live there for these 36.

What I tell the next person who right-clicks a site

Save the file, read the pixels, divide by the inches of the frame you already bought. If the number is near 100 and the day was after cameras made multi-megabyte JPEGs, go find the card, the disc, the inbox, or the book. Check the server in case the original was kept. If the original was deleted and the paper is gone, then talk about a modest enlargement and a smaller frame. Do not start with the enlargement. How to get a high resolution image is often how to stop using the low one you were handed by a theme.

The neighboring notes cover the cases that are not a drawer. How to make a high resolution image from low resolution is the avatar that was truly small, where layout and a new photo did the work. Upscale photo is the pass itself, for when the scavenger hunt fails. How to fix resolution of a photo is the tag and the inches, so a 3872 pixel file does not get mishandled by a program that thinks it is a tiny print. I needed all three ideas in my head and only one of them on the anniversary frames.

The 12 by 8 is on the hall wall. The lace is the camera's lace. The CD is back in the desk, and the copy is on two drives, because a disc that already has two unreadable files is not a place I want to be the only copy. Finding the master beats inventing one. Copying the master is the step I do before I am proud of finding it.