Epson Printer color is wrong

[Wayne Fox]

In an unusual situation which I've never faced before.

An identical target generated from i1Profiler was printed using APCU, via Windows on an Epson p7000 (located in Texas)  and with my Mac on my Epson p9000 .  The resulting targets are visually quite a bit different, and the resulting profiles are substantially different. I sent the targets to Texas, and both profiles created by the same person there using an i1i0 table by an experienced color management person.  Each profile produces good results on the printer it was printed on, and terrible results on the other printer.  Supposedly the profile printed on the p7000 has worked for several other printers, but a few have seen issues like mine.

The p7000 and p9000 profiles are supposed to be interchangeable and the variation between Epson printers should be very slight and perhaps indistinguishable.  So it seems one printer is  out of factory "calibration" (for lack of a better word).  A custom profile solves the problem to some extent, but if one machine is printing color this far out of spec it seems making up for it with a custom profile is like putting a bandaid on it.

What I'm trying to figure out is which machine is the problematic one.  I know my printer seems to print fine with epson profiles on epson paper, (although perhaps with a very slight color cast difference in a few tones compared to my 9900.  But certainly acceptable).  But I'm suspecting my printer is the problem one, because when I graph the two profiles in ColorThink, they are very different, and when I add an Epson profile for a similar surface paper (enhanced matt_) it seems to match pretty closely to the p7000 profile, where my printer seems offset from the other two.  Additionally the maker of the profile is a paper manufacturer and even though this is a very new media it seems their profile works for many users, although they did mention a few have had issues like I have.

The attached jpeg shows the 3 profiles, my profile is solid, the yellow wireframe is the p7000 profile, and the red wireframe is the Epson p7000/p9000 profile for Enhanced Matt.  I've also attached a side by side of two prints, one with my profile, one with the profile from the p9000 (an iPhone snap taken with LR mobile on an iPhone 7+ in dng and balanced so it shows a pretty accurate idea of the dramatic difference.)

I certainly would like to get Epson involved here, but want to make sure my machine is the outlier.  Seems a logical step is printing a known file such as Andrew's or Bill's on a few different Epsons, but unfortunately I don't have another p7000/p9000 local to run some comparisons on. Also,I'm puzzled as to how an Epson could be this far off, but I'm not sure how they are calibrated at factory to make them consistent from machine to machine, and if that can also be done in the field by a tech.

Any thoughts/suggestions appreciated.

[Mark D Segal]

Wayne, it may not be the printers; it could be the profiles. If you haven't done this yet and you have access to both a 7000 and a 9000 and a printer test target with patches that have known file reference values, print that target in each printer with its bespoke custom profile for the printer/paper combination, measure the printed patches with your i1Pro assuming you have one, and compare the measured values with the file values to see whether there are major differences. Also do the same for Epson's canned profiles. Compare whose profiles produce more accurate results. If there are major differences between profile outcomes and file values versus printed values for the patches, this could well be a primary indication of a profile problem. For example, a possibly important factor could be the treatment of OBAs in the profiling. Are you sure the person who made the profiles treated the "M" factor correctly? If you can, profiles made for an OBA-free paper may help sort that one out.

[Wayne Fox]

Thanks.  I guess my logic is a little different here.  It would seem that if the problem were the profile because they were created with different settings, that profile would be invalid and problematic on both machines.  The fact that a profile created from a target from one machine doesn't work for the other, and the fact that the two targets printed exactly the same with the same settings are visually obviously different (I didn't see them but talking to the paper manufacturer he said saturation and densities were very different).  This means what is coming out of the two machines without any color management is substantially different.  Custom profiling can resolve it, but the two printers are not printing similarly at all, which means one of them has some kind of issue.

I can't think of any printer settings that might affect this, but we went over all of the settings in the Epson dialog.  The only thing I can imagine might be going on is some obscure windows/mac conflict, or more likely one machine is printing differently ...

[Royce Howland]

Wayne, from your description the two main variables I see that you haven't controlled for are: 1) two different people printing the profile targets (with the possibility of some kind of workflow difference leaking in), and 2) Windows vs. Mac as the machines printing the targets. I agree that it's not likely a profile problem if the non-colour managed targets are coming out visually different right at the start of the process. If both of the variables I mentioned are not the source of any difference, then one printer is not printing to spec.

Your logic of checking profiles 3-way in ColorThink seems sounds to me. You could tighten it up by custom profiling an Epson paper in the two machines, and then checking both profiles against Epson's own generic profile. Assuming the workflow is perfect in both cases, if one of the custom profiles doesn't match the other two profiles, that would seem to be the smoking gun pointing at one of the printers.

[Doug Gray]

In an unusual situation which I've never faced before.

An identical target generated from i1Profiler was printed using APCU, via Windows on an Epson p7000 (located in Texas)  and with my Mac on my Epson p9000 .  The resulting targets are visually quite a bit different, and the resulting profiles are substantially different. I sent the targets to Texas, and both profiles created by the same person there using an i1i0 table by an experienced color management person.  Each profile produces good results on the printer it was printed on, and terrible results on the other printer.  Supposedly the profile printed on the p7000 has worked for several other printers, but a few have seen issues like mine.

Any thoughts/suggestions appreciated.

Since the two printed targets are visually different the profiles would be as well. Looking at profiles that should be different only will tell you they are different. The issue is why are the two targets from the 7000 and 9000 different?

That the printers differ that much is extremely unlikely. I would first check and make sure the printer settings are exactly the same. Then I would try to borrow a windows PC or laptop and print the target using APCU on the p9000. If the targets printed on the Mac differ from those printed on a PC you may have the problem that showed up a few months back with an IOS update. It caused a significant print color shift.

[Mark D Segal]

If the targets printed on the Mac differ from those printed on a PC you may have the problem that showed up a few months back with an IOS update. It caused a significant print color shift.

Wayne - re your post, OK, I better understand now where you are coming from - I was proposing a process of elimination one step at a time, but that may not be necessary. It would seem to be one of the printers provided you are really certain the same settings and procedure were used for making all the profiles. Also in support of this - Epson's profiles are the same for the 7000 and 9000 - they are named accordingly.

Doug, I think if the profiling targets were printed from ACPU the OS problem we experienced several months ago should not be affecting the targets because ACPU was not amended and it bypasses colour management. The OS problem would show up using those profiles to make printer test images and prints of real-world photos from a defective LR/PS version.

[Doug Gray]

Wayne - re your post, OK, I better understand now where you are coming from - I was proposing a process of elimination one step at a time, but that may not be necessary. It would seem to be one of the printers provided you are really certain the same settings and procedure were used for making all the profiles. Also in support of this - Epson's profiles are the same for the 7000 and 9000 - they are named accordingly.

Doug, I think if the profiling targets were printed from ACPU the OS problem we experienced several months ago should not be affecting the targets because ACPU was not amended and it bypasses colour management. The OS problem would show up using those profiles to make printer test images and prints of real-world photos from a defective OS version.

Mark, you are absolutely right! That didn't affect ACPU.

However, I've gotten increasingly concerned that something strange is being done in the OS's based on the increasing warnings that Photoshop gives when assigning a converted to printer device image. They pop up a window that warns one should use ACPU for bypassing color management when printing targets. I print my targets on roll paper by arranging multiple pages on the larger format in Photoshop. Then I assign an arbitrary printer profile to the target composite and print it. I scan these to create custom profiles.

My assumption is that this will produce the same driver results as ACPU but, because of the warnings, I have also started verifying by printing target sheets directly from I1Profiler and comparing them to scans of prints made from the composites to make sure the dE spreads (there are always small variations) are those normally expected. To date I have never seen a measurable, statistically significant, difference in targets printed this way v directly from I1Profiler.

However, I only use Windows. Over the years I've migrated from XP to Win7 to Win10 and with each change also check to make sure the targets produced are unchanged. So far they have been. But what I don't get is the rationale for the Photoshop warnings. I have some concern that the data path for IOS may be different, or different between versions and this is the reason for Adobe's position.

In any case the fact the printed targets differ means one of only two possibilities. Either the two printers have a really large difference indicating probable error in one of them, or the IOS path differs from the Windows path for some obscure reason.

The one way to clearly determine which is the case is to print the target using the same OS and ACPPU on both printers. If they continue to differ then at least one printer is whacked. Comparing the generated profiles with a canned one should be sufficient to determine which printer is off.

Your point about the "M" factor is certainly another variable when comparing profiles but won't result in printed targets that are visually different unless the paper used also differed. Presumably that didn't happen.

[Mark D Segal]

Hi Doug; small correction - where I said defective OS version above I meant defective LR/PS version. Anyhow the point is the same.

Yes I agree - the point in your pen-ultimate paragraph would be a good diagnostic.

I am interested in your observation about printing profiling targets from i1Profiler. Whereas in principle it should work, it has produced somewhat less accurate profiles for me compared with printing the targets from ACPU. Perhaps I should retest it.

[Doug Gray]

Hi Doug; small correction - where I said defective OS version above I meant defective LR/PS version. Anyhow the point is the same.

Yes I agree - the point in your pen-ultimate paragraph would be a good diagnostic.

I am interested in your observation about printing profiling targets from i1Profiler. Whereas in principle it should work, it has produced somewhat less accurate profiles for me compared with printing the targets from ACPU. Perhaps I should retest it.

I have yet to see any difference that I can't ascribe to the various sources of error apart from the OS printing processes. My working hypothesis is that all three approaches I've described produce the same result through the OS on Windows 10.

But I do see small variations beyond instrument ones. Some of the largest are due to paper and the position of a patch on it. Also, whether I have run a job just prior or the printer has been idle for a day or two. These dominate but there could still be differences these larger effects mask.

This gets into the interesting question of how to define "accurate."  My primary concern is that if I ask a printer to print a specific, in gamut, Lab color that the printer will do so. Generally, I have a patch set of Lab colors generated by PatchTool that are all selected to be within gamut and round-trip less than .2 dE. I then make a 16 bit tiff Lab image and print it using abs colorimetric. I scan it using measure chart. Save the CGATS. Then process it in Matlab to report the statistics. My expectation is that 90% or so of the patches will be less than 1.0 dE.

So I rate the "accuracy" of a profile based on those statistics.

However, I only have an I1Pro 2 which requires multiple passes to get the patch read consistency rms below .4 dE. It gets to be tedious. I'm kicking around the idea of getting an ISIS 2 to make this process faster and reduce instrument noise error.

[Mark D Segal]

OK, we're on about the same page re "accuracy" - the printed values should be close to the reference file values.

I use a specially configured GMCC 24-patch set (plus a more detailed grayscale), print it in ABSCOL and read it with my i1Pro2. I save the readings (M0, M1, M2) as text files which I import into a specially configured Excel spreadsheet that calculates the dE for each patch and the average for the page. A good outcome should have an average dE in the range of 0.6~2.0. Never had a result below this range, and results above it are not what I consider good. While this isn't a large sample, it's performed repeatedly as a very reliable indicator of what to expect from the printer/profile combination when making prints.

[Doug Gray]

OK, we're on about the same page re "accuracy" - the printed values should be close to the reference file values.

I use a specially configured GMCC 24-patch set (plus a more detailed grayscale), print it in ABSCOL and read it with my i1Pro2. I save the readings (M0, M1, M2) as text files which I import into a specially configured Excel spreadsheet that calculates the dE for each patch and the average for the page. A good outcome should have an average dE in the range of 0.6~2.0. Never had a result below this range, and results above it are not what I consider good. While this isn't a large sample, it's performed repeatedly as a very reliable indicator of what to expect from the printer/profile combination when making prints.

I really like Matlab for these sorts of things. For instance the following reads in two CGATS files and produces a histogram of the error distribution:

p1=ReadCgats('C:\ProgramData\X-Rite\i1Profiler\ColorSpaceRGB\Measurements\CCRef.txt');
p2=ReadCgats('C:\ProgramData\X-Rite\i1Profiler\ColorSpaceRGB\Measurements\CC1.txt');
histogram(deltaE76(p1(:,7:9), p2(:,7:9))  % columns 7:9 contain Lab values

I consider an average dE of 1 to be reasonable. That's also the range I get as well. For folks w/o a spectro, printing a Colorchecker image in Abs Colorimetry is a good way for people to quickly check a profile visually against an actual colorchecker.

I'm running a target print check with ACPU, I1Profiler direct, and my Photoshop assign workaround and should have the results when the paper has dried a few hours.

[Mark D Segal]

I'm running a target print check with ACPU, I1Profiler direct, and my Photoshop assign workaround and should have the results when the paper has dried a few hours.

Looking forward to read of your results.

[Doug Gray]

I printed 4 charts with 283 patches. Two were directly from I1Profiler, one from the Photoshop workaround, and one from ACPU.

The results indicated a typical dE of .7 so between each of them with most of the variation associated with highly saturated colors.

I averaged the 4 sets then compared each set to that average.

I1Profiler sample 1: Average dE .52
I1Profiler sample 2: Average dE .27
Photoshop workaround:  Ave dE .47
ACPU: Average dE .37

No obvious difference beyond the usual variation I typically see with scans.


As an aside, when I first ran them the Photoshop version had an average dE of 19. I realized I had selected "let printer manage color" by mistake. Duh. I should have realized the settings were wrong when I didn't get the pop-up that said: "It appears you are trying to print without color management"  This occurs whenever the printer profile you select is the same one that the image has applied to it. Just cancel and you are good to go. Reprinted and scanned.

If I did that I'm sure others make the same error and when you send them off to get a custom profile things don't work so well and Adobe, amongst others, gets called. That's possibly why they have officially dropped the option. But where there's a will there's a way. At least sometimes.

[Mark D Segal]

Interesting, thanks.

[eronald]

There's an Epson calibration utility, I believe, which can be used with an i1 to calibrate the printer hardware.
I forget what it's called.

Edmund

[unesco]

There's an Epson calibration utility, I believe, which can be used with an i1 to calibrate the printer hardware.
I forget what it's called.

ColorBase and ColorBase2

[Mark D Segal]

I printed 4 charts with 283 patches. Two were directly from I1Profiler, one from the Photoshop workaround, and one from ACPU.

The results indicated a typical dE of .7 so between each of them with most of the variation associated with highly saturated colors.

I averaged the 4 sets then compared each set to that average.

I1Profiler sample 1: Average dE .52
I1Profiler sample 2: Average dE .27
Photoshop workaround:  Ave dE .47
ACPU: Average dE .37

No obvious difference beyond the usual variation I typically see with scans.


As an aside, when I first ran them the Photoshop version had an average dE of 19. I realized I had selected "let printer manage color" by mistake. Duh. I should have realized the settings were wrong when I didn't get the pop-up that said: "It appears you are trying to print without color management"  This occurs whenever the printer profile you select is the same one that the image has applied to it. Just cancel and you are good to go. Reprinted and scanned.

If I did that I'm sure others make the same error and when you send them off to get a custom profile things don't work so well and Adobe, amongst others, gets called. That's possibly why they have officially dropped the option. But where there's a will there's a way. At least sometimes.

Doug, looking back through this thread I didn't see you mentioning which printer/paper combination or combinations you were using for these tests. Grateful if you could let us know what you used.

[Doug Gray]

Mark,

All these tests were done with a Canon 9500 II and Costco Glossy paper. I've seen similar results on the Epson 9800 with Costco glossy. I haven't run ACPU nor I1Profiler directly on the 9800 with roll paper, only the Photoshop workaround.

[Mark D Segal]

OK, thanks Doug.

[Wayne Fox]

The manufacturer informed me today that some other users were having issues with their profile on this new media and reported that the profile they made from my printer was much better and produced much better results.  So it appears that either their p7000 has an issue or there was some other issue which is on their end.

We both printed targets using the same files and ACPU, mine from a Mac and his from Windows.  His target was visually quite different from mine, but produced a profile which works for his p7000, which means their is something funky with the way windows color management is working on his setup, or his printer is out of factory spec.

At this point I guess I won't worry about it, since the profile is performing quite well, and all other medias I use I make my own profiles. This particular media is quite thick so I couldn't use my iSis to make my own profile.

I appreciate all who offered thoughts here... LuLa is the only place I really come to for this type of stuff.  Lots of great minds hanging out ...