WHERE 16-BIT CAN BE BETTER.
A 16-bit file can have very minute differences between pixels--1/256th of the
minimum difference in an 8-bit file. Anything that small will have no impact on
the final reproduction--no possible sequence of editing events could ever
create a variation that anybody could see. The maximum initial difference
between a pixel of an 8-bit and a 16-bit file would be slightly less than half a
level. That is, in a 16-bit file there might be a value of 128.49, which would be
treated as a value of 128.00 in 8-bit. That half-level difference won't do
anything, either, *unless* some unlikely sequence of commands drives it
much, much further away from where it would be if it were an 8-bit file.

For technical reasons that will be discussed later (see What the Extra Bits
Actually Do, Part III), if you look hard enough and at a high enough
magnification, the 16-bit edit always looks marginally smoother and the 8-bit
more active. To date, I know of four types of natural photographs that, if edited
to an extreme, show differences large enough for people to prefer one or the
other. (There are also some times in retouching and image conversions
where working in 16-bit helps, but they are so esoteric that I have rarely
written about them.)

1) If we apply massive edits to a grayscale file, the difference between an 8-bit
and a 16-bit correction may become noticeable. The 16-bit version would be
preferred if the image featured areas where smoothness is desirable, like
skies; the 8-bit when the subject is full of detail. In my testing, even with very
big grayscale edits, well over half of the images showed no difference. Of the
others, the result of 8-bit editing was preferred roughly twice as often as 16-bit
edits. But definitely 16-bit editing got better results in certain images. The
reason that this does not carry over into color images is that when three
channels are superimposed on one another any variation in one is less
visible. Similar massive edits to the RGB files that were the ancestors of the
grayscale files showed no difference of any consequence.

2) Early in my testing, one list member provided a demonstration based on
applying the same edits to one 16-bit and one 8-bit file, both generated by a
scanner from a single scan. I verified that when the edits were applied the 8-
bit file looked distinctly worse. However, when the tests were repeated on a
copy of the 16-bit file that had been converted into 8-bit not by the scanner but
in Photoshop, there was no difference in quality. I communicated this finding
to the list in 2001 and recommended that people take 16-bit files from
scanners where possible.

3) A second user provided a similar exercise where edits were applied to 16-
bit and 8-bit files generated in Canon acquire software from the same digital
capture. (The original had deliberately been acquired incorrectly in order to
make the differences more apparent, which disqualifies it as a real-world
example, but in view of the interesting nature of the problem I followed
through with the testing.) Again, I verified that there was a quality loss by
editing the user's 8-bit file, again I retested by converting his 16-bit file to 8-bit
in Photoshop, and editing that. As with Example #2 there was now no longer
a quality difference, so I recommended to the list that we avoid taking 8-bit
files directly from a camera package when a 16-bit file is available. I do not
know whether the same problem exists in Camera Raw but I will be testing it
in coming months.

4) In 2005, a third user provided a Camera Raw file of a scene of a city at
night. He sabotaged the image by moving the exposure slider within Camera
Raw all the way to the left in spite of the fact that the image was already too
dark. Then, he acquired the image in ProPhoto RGB, an ultra-wide gamut
RGB definition that is rarely used in professional work. The image contained a
large area of sky. Applying the drastic curves that were needed to lighten the
image to the 16-bit file resulted in a perceptibly smoother and more attractive
sky than the one done by converting the file to 8-bit in Photoshop and
applying the same curves there. When the same image was captured with the
same sabotage in Camera Raw into either of the RGB definitions that most of
us use--the narrow-gamut sRGB or the wide-gamut Adobe RGB--there was
no significant difference between correcting in 8-bit or 16-bit.

THE RETREAT.
Ever since the initial assertions that 16-bit editing would create an enormous
difference, its proponents have been in full retreat as users have asked them
again and again for any example to support the notion. They have provided a
blizzard of gradients and histograms, but never a real image. One author's
idea of illustrating the concept was to compare an *original* image to one that
had been edited in 8-bit and then showing the histogram. In a second book,
he compared reasonable editing in Camera Raw (which is 16-bit) with idiotic
editing in 8-bit Photoshop, in each case claiming that it showed the superiority
of 16-bit editing.

In fairness, the demand for images placed the advocates in a difficult position.
There is no reason to doubt that they actually believed their original wild
claims were true; as has been made abundantly clear since, they never
bothered to run tests before making them. By the time they learned that there
was serious doubt that there was any 16-bit benefit at all, let alone a night and
day difference, they had already begun to promote seminars about the
benefits of the 16-bit workflow. Furthermore, Adobe, largely at their
suggestion, had begun to add 16-bit capabilities to Photoshop which were
being heavily hyped. As many of these advocates take money or other
support from Adobe, it would have been exceedingly awkward if they had
abandoned the you-are-not-a-professional-if-you-don't-use-16-bit line.

Since they could not abandon their position but could not produce anything to
back it up, they resorted to smokescreens. The usual method was personal
attacks on me. They repeatedly referred to some mysterious "agenda" of
mine. They called me lots of names, but never could get around to showing
what the people were asking for. They asserted that I said 16-bit was
worthless under all circumstances and presented gradient after histogram to
prove that it wasn't. They spent scores of hours telling users that they were
"too busy" to prepare demonstration images that could have been made very
quickly if the difference was even a tenth as critical as what they were touting.
They constantly tried to evade responsibility by saying that the burden was on
me to prove that 16-bit doesn't have advantages, as if *I* were the one who
was saying that anyone who didn't work my way was unprofessional and *
they* were the ones who were tolerating either way.

One of the more prominent advocates, Bruce Lindbloom, was so frustrated by
his inability to produce a persuasive image that he posted a web page that
accused me of sabotaging my 16-bit images before testing them. Also, he
asserted that I kept my results private and that nobody else could verify them.
Both statements are categorically false, and Lindbloom knew that they were
false when he posted them. 16-bit advocates Andrew Rodney and, to a lesser
extent, Bruce Fraser, both of whom are well aware that the Lindbloom page is
a crock from the word go, nevertheless repeatedly post links to it, hoping that
if they post the falsehood enough times, it will magically become true.

The gyrations that these advocates went through to explain why they could
not produce even a single real image that would support the notion that 16-bit
editing was "highly critical", would produce a "night and day difference", and
so forth are so remarkable that they are excerpted at length in Part IV.

This list did get a glimpse of the rationale from Jeff Schewe in 2002, in a
thread that is posted in our archives. He defended his assertion that those
using 8-bit are "recreational, rather than professional" users of Photoshop as
follows (note: Jeff frequently uses ellipses [. . .] in his messages; in the one
case where I have deleted an extraneous section of the message, I use ***):

"Nope. . .Dan and the rest of you are welcome to continue scanning in 8 bits
and doing whatever you want to do to your images. . .but if you want absolute
total control over tone and color without the risk of breaking the image
somewhere down the road. . .you better learn to edit in 16 bits.***And yes, I'll
stand by the line 'recreational' if you squander and waste your data bits just
getting an image tone/color corrected in 8 bit. . .cause if you do that, you're
working with considerably less than 8 bits/channel and deserve the banding
you are likely to incur."

I replied, "Rather than continuing to post the same defensive bluster to every
group credulous enough to listen, it must be a better use of your time to
produce even one image that demonstrates the point. After all, this is
supposed to be critical, night and day, the difference between professional
and recreational imaging. If an image exists that shows such a dramatic
difference, why not show it, rather than just make claims?"

The evidence Jeff offered in response was, "Pretty much all of my work the
last 5 years was scanned in 16 bit for initial tone & color correction. You are
welcome to look at my work and see for yourself, no banding. . .even after
hours of editing and extreme manipulations. I'll let my work speak for itself."

Within a year, however, others had backed off the original claims. The new
emphasis was on "flexibility for the future." The night and day differences were
no longer found in really big edits, but only in ones with multiple big edits. We
were cautioned that, even if 8-bit is sufficient now, new types of output devices
might arise that would require more bits. The ad hominem attacks on me and
my purported motivations continued whenever users asked on-line for
specific images.

By 2004 the embarrassment was such that the protagonists began to deny
ever having made any of the apocalyptic statements. In November, Andrew
Rodney posted the following astonishing statement to this list: "And no I've not
seen any text that says '16-bit is absolutely critical, creates a night and day
difference, that anyone who doesn't do it is an amateur, etc., etc'. It's simply a
reflection of math and physics." When Ric Cohn pointed out that Andrew's
own business partners had frequently used precisely those words, the
following day Andrew denied having denied it. Thereupon I produced
Andrew's original denial--and he denied ever denying he had denied it. And
throughout the rest of the thread he resumed the position that nobody had
ever said such things. And held to it, even after the quotes were posted.

Earlier this year, on the ColorSync list, of which I am a member but don't
normally participate, Bruce Fraser once again took shots at my motivations
when the list turned to the bit depth topic. He wrote, "What Dan's tedious and
fundamentally specious arguments deliberately miss is that the need for
greater bit depth has absolutely nothing to do with reproduction and
everything to do with editability."

At that point I entered the thread to point out that I did not miss that point and
that on the contrary, the files that I had been testing were edited beyond all
recognition, beyond any possible claim of real-world practice. And I pointed
out how many other people had performed similar tests with the same results,
and again asked why he could not produce any images to back up his claims
of a "night and day difference" that was "totally obvious to anyone who looks".

Bruce denied having ever said these things, whereupon I produced the
original files where he did say them. After further exchanges in which he
waffled somewhat on these phrases, I asked pointblank whether he had ever
personally run tests of 8-bit editing vs. 16-bit editing, before or after having
laid down these ukases about how the difference would be night and day. He
refused to answer. After being further pressed, he stated that it would be a
waste of time because it was obvious that the 16-bit would look much better (if
you don't believe he said this, don't take my word for it, go to Part IV). And,
having refused to accept the possibility that something he hadn't tested might
not be true, he ended with a typical slur: "My personal opinion is that this is a
manufactured controversy--I decline to speculate on the motivation of those
who have manufactured it--and I'm utterly disinclined to waste my time
arguing the point when I have better things to do with it."

In many of these threads, other users have chimed in claiming that they are
positive from first-hand knowledge that 16-bit editing avoids problems of
banding. I'm aware of around thirty such posts, including a couple to this list.
In about a dozen cases, I've gone off-line to ask these people how they are so
sure. Without exception, they have never performed any testing--it's all a
hunch. They, like Jeff and Bruce, merely are supremely confident that their
work would show banding or other artifacting if they did it in 8-bit. But they've
never tried to do the same things to 8-bit files that they do to their 16-bit files,
so they can't know for sure, and they refuse to accept the word of others who *
have* performed such tests.

The people who have doubted that there is a night-and-day difference
between 16-bit and 8-bit correction, or that correcting in 8-bit is amateurish,
have generally said they are willing to be persuaded otherwise by examples,
as I am. Some members of the other side take a different tack. It is so obvious
to them that their view is right that not only do they require no proof of it, but
they state outright that they refuse to accept any proof that it is not. It has been
remarked by others that such a position is a religious rather than a rational
one, and I agree. The closest analogy I can think of would be to the person
who says that it is so obvious that the world is flat that it needs no proof, and
that any demonstrations that it is round will be ignored because they can't
possibly be right, since it is well known that the world is flat.

It actually gets better. Since that time, Bruce Fraser has, incredibly,
announced that direct comparison of 8-bit to 16-bit editing is invalid, and
conceded that side-by-side tests will show no advantage for 16-bit. He writes,
"I've demonstrated many times things that work better in 16-bit than in 8 bit,
but Dan has rejected these because they don't fit his narrow criterion of doing
exactly the same things to a 16-bit and an 8-bit file, then comparing the
results." Separately, he clarifies, "The major problem with the methodology...is
that by making identical edits to the 8-bit and the 16-bit, you're throwing out
any benefit the extra bits may bring. They aren't useful unless you DO
SOMETHING with them!"

The extra bits may indeed be useful if you do something, but there's no way of
knowing for sure without trying to do the something without them, and seeing
if there's a significant difference. A lot of people have done this. The answer
they have unanimously come up with is that there is not.

Dan Margulis

[prev] [next]