To flatten or not to flatten before Resizing/Resampling

[chris_ould]

<p>Having soft-proofed a duplicate of your master image and added a couple of adjustment layers to correct for the output process, is it best practice to flatten before using the Image Size dialog box to Resample or Resize the image? Or does it not really matter?<br>

Thanks</p>

[digitaldog]

<p>Need more info on the workflow. You have an output ready image with soft proof adjustments. Are you printing this yourself or sending off for output? Will you need to further edit the image (with those proof adjustments)? You can go either way, it will be a bit slower if you keep all the layers but for future flexibility you may want to keep the layers. If you are sending them off to a service provider, flatten! </p>

[sarah_fox]

<p>If it's an image you care about, save a full-resolution, master working file with all your layers, etc. Then save to another merged output file for printing or displaying. If you ever need to do anything different with your image, at least all your work will be preserved in the master file.</p>

[chris_ould]

<p>Regarding the workflow, i would not need to edit the image any further. The next step after Resizing/Resampling would be to apply output sharpening before sending it to my printer (or occasionally converting it to a profile provided by a print lab before sending it to them).<br>

I was just curious if flattening it before Resizing/Resampling was strictly necessary to avoid any artifacts etc.?<br>

Thanks for your responses.</p>

[digitaldog]

<p>Based on what you've written, I'd flatten. This is an output ready file sized, sharpened etc for this one device so extra layers buy you nothing. </p>

<blockquote>

<p>I was just curious if flattening it before Resizing/Resampling was strictly necessary to avoid any artifacts etc.?</p>

</blockquote>

<p>It's going to be '<em>flattened</em>' at some point anyway. If you print an image in-house, through your own driver, that layered image <strong>is</strong> flattened as it goes to the print driver. That is why the idea that layers are <em>non destructive</em> is not fully true. Yes you have editing flexibility! But no, those edits ARE applied as if you flattened the data at some point (like when you print). So the rounding errors and anything that could apply data loss and possible artifacts will be applied to the data. No way around that (expect doing this all with parametric edits on raw data). </p>

[EricM]

<blockquote>

<p>I was just curious if flattening it before Resizing/Resampling was strictly necessary to avoid any artifacts etc.?</p>

</blockquote>

<p> <br>

No. You can resize with layers/smart objects without causing artifacts etc. Sarah offers great advice and is how most work. Save your layered master file. Then flatten it and save it with a suffix like "fltn", do your resizing for your desired print, do your output sharpening, convert to sRGB, convert from 16 to 8 bit, save as jpg or tif and upload to Costco etc.</p>

 

[digitaldog]

<p>The OP clearly understands his workflow and doesn't need to be told to save a layered master file as he writes quite clearly:</p>

<blockquote>

<p>Having soft-proofed a <strong>duplicate</strong> of your <strong>master image</strong>....</p>

</blockquote>

<p>The copy is edited for a specific output device (soft-proofed a duplicate), it's separate from the master which is edited for the working space. The question is about flattening this iteration which is a good idea if being sent to an outside lab who's brains will explode getting said layers. Now if he's printing in house, keep those output specific layers in case there is a need to tweak (which is possible with the outside lab too but again, don't send them a layered file). <br>

Either way, if the data is such that the edits produce artifacts, it's going to occur flatted or not. The edits <strong>have</strong> to be applied for the image to be printed. </p>

[EricM]

<blockquote>

<p>is it best practice to flatten before using the Image Size dialog box to Resample or Resize the image? Or does it not really matter?</p>

</blockquote>

<p> </p>

<blockquote>

<p>I was just curious if flattening it before Resizing/Resampling was strictly necessary to avoid any artifacts etc.?</p>

</blockquote>

<p>Chris has a clear question, that he's now made in two posts, but you still haven't answered it.</p>

[digitaldog]

<p>Yes I have, I'm afraid you didn't understand the text:</p>

<blockquote>

<p>1. You can go either way, it will be a bit slower if you keep all the layers but for future flexibility you may want to keep the layers. <strong>If you are sending them off to a service provider, flatten!</strong><br>

<br /> 2.Either way, <strong>if the data is such that the edits produce artifacts, it's going to occur flatted or not.</strong> The edits have to be applied for the image to be printed.</p>

</blockquote>

[EricM]

My gosh. Argue.net. Chris simply wants to know if a

layered file gets damaged when resizing and if he/she should flatten to avoid possible artifacts.

 

You don't

need to know anything about his/her workflow, where Chris

prints, or what camera was used or if it's raining

[digitaldog]

<blockquote>

<p>My gosh. Argue.net. Chris simply wants to know if a layered file gets damaged when resizing and if he/she should flatten to avoid possible artifacts.</p>

</blockquote>

<p>And Chris got the answer, nothing to argue about. </p>

<blockquote>

<p>You don't need to know anything about his/her workflow, where Chris prints, or what camera was used or if it's raining</p>

</blockquote>

<p>If you want to make assumptions to answer questions, then yes, you don't need to know about workflow. IF you wish to answer the question correctly and aid the OP, knowing his workflow is useful. <br>

<br>

If you actually <em>read</em> what the OP asks, (and understand the text), then it's unnecessary to discuss <em>master files</em> when it's quite clear Chris knows the difference between one and the iteration that he clearly created <strong>for output</strong>. <br>

<br>

Sending a layered file to an outside provider only ensures sending a larger file, one that has layers that could be affected (which he probably doesn't want) and complicates a workflow (of which <em>some</em> suggest isn't pertinent). </p>

[EricM]

<blockquote>

<p>If you want to make assumptions to answer questions, then yes, you don't need to know about workflow.</p>

</blockquote>

<p><br /> If workflow is important, like you claim, you should be able to enlighten us with <strong>a workflow that would cause artifacts with a layered photoshop file when re-sizing?</strong> <br /> <br /> Re-sizing an image with multiple layers, or flattened, wont cause artifacts. What kind of workflow leading up to, or after, re-sizing, is irrelevant to this. Excuse the noise, Chris.<br /> </p>

[digitaldog]

<blockquote>

<p>If workflow is important, like you claim, you should be able to enlighten us with <strong>a workflow that would cause artifacts with a layered photoshop file when re-sizing</strong></p>

</blockquote>

<p><strong><br /></strong>Love to. Once you start a <strong>new</strong> post with that question. Until we hear back from Chris, there's little reason to continue going OT and hijacking the thread about the easy processing routines that can cause severe data loss with or without layers, before or after resizing! <br>

Before doing so, start your education on the subject by reading up on the subject of data loss that can result in artifacts (banding, posterization etc):<br>

http://www.digitalphotopro.com/gear/imaging-tech/the-bit-depth-decision.html</p>

<blockquote>

<p>Re-sizing an image with multiple layers, or flattened, wont cause artifacts.</p>

</blockquote>

<p>No one said that <strong>alone</strong> would cause artifacts. You <strong>continue</strong> to have difficulty reading and understanding the text already presented:</p>

<blockquote>

<p><em>But no, those edits ARE applied as if you flattened the data at some point (like when you print). So the rounding errors and <strong>anything that could apply data loss and possible artifacts</strong> will be applied to the data. No way around that (expect doing this all with parametric edits on raw data).</em></p>

</blockquote>

<p>and<em>:</em></p>

<blockquote>

<p><em>Either way, <strong>if the data is such that the edits produce artifacts, it's going to occur flatted or not.</strong> The edits have to be applied for the image to be printed.</em></p>

</blockquote>

[john_wheeler6]

<p>Hi Chris<br>

IMHO (and experience) it is possible to create more anomalies if you resize before flattening. <br>

In addition, so you are not viewing (or missing) artificial artifacts, best to do your viewing at 100% (even more important if you have not flattened). PS and the display driver can both introduce artificial artifacts viewing Layered images at non-100% under various conditions.<br />Some simple example images to demonstrate later this evening.</p>

[digitaldog]

<p>John is absolutely correct about viewing at 100%! What may appear is '<em>artifacts</em>' don't actually exist in the document. A zoomed out view can show what appears to be but isn't data loss. <br>

In <em>some</em> cases, sampling down can reduce noise and the algorithm plays a role. Noise is found in very dark areas of an image. As you resample down, single pixels of non image forming '<em>data</em>' (noise) is removed when replaced with actual dark pixels adjacent to that pixel. <br>

Resampling a flattened document should be faster than resampling a layered document so if you know you're sending the lab a flattened file to print, makes sense to flatten then sample. That's workflow, not work-slow.<br>

Looking forward to seeing John's example. But again, it is rather pointless to send a layered file to a print provider and since you must flatten, do it first, then resample. Should be faster. </p>

[john_wheeler6]

<p>Andrews comments lead right into the example below.<br>

<br /> Just one source of anomalies that can be created by resizing fist is that Photoshop resizes the pixels on each Layer by itself as well as the Layer mask by themselves. When high contrast adjacent pixels exist, they are merged into closer to a gray. How that gray operates in masks or in Layer blends is totally different than if you flatten first.<br>

<br /> I have wanted to create this simple example below for some time so this was a good opportunity to complete the project. It shows the same file on the left as on the right with the exception that the left side was flattened that resized by 50% and the right size was resized by 50% and then flattened. The left side gives the correct rendition every time.<br>

<br /> The base image is created with a horizontal spectrum gradient.<br /> The second Layer is simply a pixel checkerboard of black and white pixels.<br /> I duplicated the upper layer 27 times, masked a horizontal slice and assigned a different blend to each of those duplicated Layers.<br>

<br /> When you resize first, the black and white checkerboard turns into a solid gray first. This totally changes the blend compared to adjacent black and white pixels in most cases.<br>

<br /> The example below does not cover all cases of blends as not all blends are commutative by Layer. Also, just because one type of blend looks like it works resizing first, I assure you, I can craft an example where it will not work right.<br /> Even though this is a contrived worst case example, it graphically shows how high contrast and high frequency content can create a huge difference between the resizing first vs flattening first.<br>

<br /> The image is large and scaled down by photo.net, so if you want the 100% view use this link: http://jkwphoto.smugmug.com/Other/Photonet/41170890_tSbB5W#!i=3839141482&k=gkVDpsT&lb=1&s=O<br /> <br /> The image below is somewhat shrunk by photo.net yet should show the problem with resizing first:<br /> <img src="http://jkwphoto.smugmug.com/photos/i-gkVDpsT/0/O/i-gkVDpsT.png" alt="" width="840" height="810" /></p>

[Uhooru]

<p>To answer original question, I don't think it really matters except it sometimes when you resize if down-sizing, depending on how much you do so, dark areas can appear darker. Maybe then, it would be better to flatten after you resize just so you can make any needed final adjustments in your layers at final size. </p>

[chris_ould]

<p>Thank's for such concise responses to my original question.<br>

I had always considered it 'standard' practice to flatten before resizing/resampling during a print workflow, but had forgotten where i got that information from (possibly Jeff Schewe). <br>

Anyway, now i know why! <br>

Best Wishes<br>

Chris</p>

 

[john_wheeler6]

<p>You're welcome Chris<br /> <br /> And as a second source of anomalies after flattening first is which rendering choice is made in the Image > Image Resize command.<br /> <br /> Using the same image as in my prior post, the left image is flattening then resizing with automatic rendering (which I used in my prior post) and the one on the right is doing the identical steps except choosing nearest neighbor rendering.</p>

<p>The example was created to maximize artifacts so they were clearly visible. Most images will not have any noticeable artifacts or those that exist can be easily overlooked (unless you make a large print). Knowing the better default approach can help one stay out of trouble on those occasions when an image is prone to the creation of artifacts by PS manipulation.</p>

<p>So "resizing" does matter :)</p>

<p>Resizing with fine pattern high contrast information in the image does not always work well with the nearest neighbor rendering. Automatic works well yet if you find suspected anomalies, you can always try another rendering option.<br /> <br /> <br />To see the 100% view right click on image and download or view on your local viewer<br /> <img src="http://jkwphoto.smugmug.com/photos/i-jTnWf89/0/O/i-jTnWf89.png" alt="" width="660" height="810" /></p>

[digitaldog]

<p>Seems Adobe needs to add a <em>Flatten</em> check box to<em> Image Size</em> as we see in the <em>Convert to Profile</em> command (which really <strong>can</strong> play a role here indicating that workflow IS important when converting and as well, sizing). <br>

Another test, one that's simple is to create a four pixel PS document. Make three pixels 0/0/0 and the forth 128/128/128 and without any layers, resample down 50% and examine the results of the final pixel value. As expected, Nearest Neighbor plays a profound role in the resulting value compared to other algorithm's. Alter the values slightly (0/1/0, 1/0/0, 0/0/1) for the three pixels. <br>

Sound workflow dictates (again) it's kind of silly to send anyone who isn't editing the data a layered document simply for output. The document has been resized for that output! And considering that a sound workflow suggests output sharpening at output resolution, sizing is nearly 2nd to last step for processing this document so it's pointless to have layers at this junction. Flatten, size, output sharpen. Convert to the output color space <strong>if</strong> allowed and flatten at that stage since it's an available option.</p>

[sarah_fox]

<p>John, very interesting! Thanks for doing that experiment for us! Fortunately I've always created a flattened output file from my master file for resizing, output sharpening, etc., but I can't say I knew it could make that much of a difference. I had assumed it wouldn't really make much difference when the resizing was done. ;-)</p>

[john_houghton]

<p>There is one particular situation for which it is important to flatten before resizing and it really does matter. This concerns images covering a full 360 degrees when the left hand edge must exactly match the right hand edge at the wrap join when viewed in a spherical viewer (as in Google street view for example). If you resize before flattening, you will as often as not get a visible pale vertical line at the join where the images butt together. The line is formed from lighter versions of the edge pixels.</p>

[EricM]

<p>Thank you John for that effort and your conclusion. That's fantastic info with a definitive answer. It's the first anyone here has heard, I'm sure. I hope the info spreads as it's not common knowledge<br>

<br>

</p>

<blockquote>

<p> <br>

Just one source of anomalies that can be created by resizing fist is that Photoshop resizes the pixels on each Layer by itself as well as the Layer mask by themselves.</p>

</blockquote>

<p> <br>

No contest, that's a common sense. However there is a whole industry of teachers, graphic designers, retouchers, composers, and editors that collaborate on layered psd files that go in and out of photoshop, premier, illustrator, indesign that resize multi-layered psd and tif files as an industry practice that have never complained...or noticed :) <br>

<br>

</p>

<blockquote>

<p>The example was created to maximize artifacts so they were clearly visible. Most images will not have any noticeable artifacts or those that exist can be easily overlooked </p>

</blockquote>

<p> <br>

Exactly. That brings me to a few questions. What extremes did you down size, and by what amount, to achieve this? Was 27 layers the amount of layers you needed before you noticed a difference? Would you be willing to replicate and share this procedure on an an actual image here? Would you be willing to share your psd file via dropbox etc so others can pixel peep at 300% and determine what the difference is on their own images?<br>

<br>

Chris, what did your real-world experiments determine before posting here?</p>

<p>Thanks again John, that was brilliant </p>

 

[chris_ould]

<p>Eric,<br>

To be honest, i was only interested in finding out if a duplicate file with a couple of adjustments layers to correct for color shifts and loss of contrast (symptoms of soft-proofing) 'really' needed flattening before resizing/resampling...and from what i can see at 100%, the difference is very marginal but after taking a look at John's experiments i think it fair to say that it does!<br>

<br>

Perhaps, as Andrew suggests, their could be an option 'to flatten' in the Image Size dialog box?</p>

 

[sarah_fox]

<p>Chris, another option for you, if you really want your output file to preserve the layers, is to use "copy merged" to copy all the merged layers at full size, and then paste the merged information to a new layer on top of all the other layers. Then when you resize, all you will see is the resizing effects on the merged layer. You will also be able to tell whether resizing had any effect on the unmerged appearance by toggling the visibility of the overlying merged layer.</p>