Comments on: Simplicity vs. Power in Photoshop

By: Chris Cox

Chris Cox — Sun, 02 Nov 2008 16:45:22 +0000

The conversion to smart objects can’t use threads for much – it’s just doing some in-memory file conversions. And the compositing/transform parts already use threads. Smart FIlters also use threads, just like normal filters.

By: Stephen Best

Stephen Best — Sun, 02 Nov 2008 00:58:25 +0000

One of the really disappointing things with PSCS4 is that “Convert to Smart Object” still seems to be single-threaded. Is there a technical reason for this or is this something Adobe just hasn’t got around to updating? I’m often converting the luminosity component of large film scans to a SO for sharpening etc and it takes forever on my 8-way Mac Pro.

By: Ted Dillard

Ted Dillard — Sat, 01 Nov 2008 10:46:17 +0000

Once again, a great read.
I’ve been teaching Smart Objects in a number of venues for since they came out… I get the resistance and confusion a lot, thus the book outlining a “foundation” workflow.
It’s funny, I usually start the class with a, “while you’re in THIS class, you’re doing it MY way”. :)
Thanks again, John!

By: ajit

ajit — Fri, 07 Dec 2007 19:37:48 +0000

I am encouraged that you are discussing this because I have been praying for a photoshop-replacement. Something with half the power but much simpler. Some of the choices in how tools work in Photoshop baffle me. The crop tool in Motion versus Photoshop highlights what could have been.
Illustrator is near unusable but I do. Reluctantly.

By: Chris Cox

Chris Cox — Fri, 05 Jan 2007 10:20:42 +0000

Adolfo – yes, it is quite a bit like AE.
We looked at a lot of existing approaches in different applications, plus a few new ideas, and weighed the advantages and problems of each.
(well, that plus lots of hair pulling, screaming in frustration, gnashing of teeth and long discussions in John’s office).

By: Chris Cox

Chris Cox — Fri, 05 Jan 2007 10:18:27 +0000

Ben – blurring and sharpening filters as adjustment layers lead to the performance problems that John already mentioned.
We’ve tried it, and it was too @#$%@#%$ painful to inflict on our customers (you should have seen the colorful comments from our alpha testers).

By: Chris Cox

Chris Cox — Fri, 05 Jan 2007 10:16:19 +0000

dd – you can already use Smart Objects with multiple layers (or other Smart Objects). Smart Objects don’t place any limit on the contents of the child document/file. If you have the layers in the parent document, just select the ones you want and Convert to Smart Object. Or you can place a file that you’ve already saved with all the layers.
Plugins: sometimes it’s much easier to test new code and new ideas in a plugin framework instead of the main application. And some of those operations require so many custom controls that having them run inside Photoshop would make a really dreadful UI. Thus, we isolate them in their own window. At that point, it almost doesn’t matter that they’re plugins….
Also, plugins are a whole lot easier to patch/dot release than the whole application. And plugins load/unload on demand, so their large chunks of code don’t impact the main application.
Warp: we just haven’t had time to increase the functionality. Getting it working right for the simple (single patch) case took longer than expected (lots and lots of edge cases).
Also, node/graph based editing confuses the heck out of photographers, printers, artists, illustrators, etc. The people who seem to grasp node based editing are the ones who have some idea of the operations going on “behind the scenes”. Yeah, we’ve tested the UI, and it didn’t fare well.

By: Adolfo Rozenfeld

Adolfo Rozenfeld — Sat, 30 Dec 2006 22:39:54 +0000

“After Effects is the model. Make it work like After Effects. Smartly cache objects when the pixels are directly edited like AE does with text layers that are converted to outlines”.
Actually, After Effects’ paint engine (the one area in which AE operates in the same playing field as PS) requires to open the layer in a separate window from the main composition. The reason for that, of course, is that it could become unbearable slow if it worked in the main comp panel, in the conext of a fairly complex composition. So smart objects are very AE-like in this regard. There are other functions in AE, like motion tracking, which also require opening the layer in a separate panel.
I don’t understand what you meant by caching text created to outlines. When AE 7 converts text to outlines it simply creates a solid, with as many masks as needed to reproduce the original vector outlines for type. No caching of any kind.

By: Ben Richardson

Ben Richardson — Sat, 30 Dec 2006 18:52:19 +0000

I just posted a comment on Shake (as mentioned above) to yesterdays post. The short line is, yes it works, and yes there are limitations. However, it made me realise that all I really want is live sharpening (and possibly blurring). Realistically, a direct Adjustment Layer style Sharpening Layer is the only thing that would directly and practically improve my workflow, and even that might be limiting (since I tend to sharpen channels separately). I appreciate the thought that's gone into Smart Objects, but I'm not too comfortable with converting my entire, HUGE multi-layered composites to Smart Objects, just to gain live sharpening. Related question: are Smart Objects really as robust as normal layered files? The extra Save step to commit changes has always given me chills... [Ben, I don't think you'll have any problems with Smart Objects. They are PSB files that live inside your PSD or PSB file. You can always choose to save that file out separately. Anyway, I hear you about the overhead tied to Smart Objects. We could offer the option not to store that data inside the saved file, but you'd have to be cool with re-rendering it (e.g. applying warping, filtering, whatever) on the fly when opening it, and you'd have to live with reduced compatibility with other apps, older versions of PS, and even copies of PS that don't have the same filters installed. I do think it's reasonable to offer the option to make that trade-off, but it's not something we could undertake for this rev. --J.]

By: Rich MacDonald

Rich MacDonald — Fri, 29 Dec 2006 20:52:24 +0000

The SO concept sounded amazing to me when first reading about CS2, but I haven’t used it much because of the transformation limitations (no perspective/distort). I was/still am hoping to see this in CS3.
[Ah, I wish I could hold out hope, but it's not planned. Its absence is a drag, I know. I will mention it one more time, but the best dude to do the work has been ill. --J.]
Using smart filters with Lens Distortion we can kind of get this, but it’s not as intuitive when trying to align multiple images.
This leads to the “Auto-Align Layers…” feature. It’s good to see this addition (and that the SIFT algorithm has made its way into PhotoShop). At the same time, as convenient and powerful as the auto functions are, we all know times when we need to more tightly control adjustments. To this end, I’m surprised that Free Transform hasn’t added a means to select the 4 adjustment points to define the homography for the transformation (i.e. instead of making us use the layer’s bounding box corners, sometimes which aren’t even above actual pixel content). This would be a killer feature. Better yet, add the ability to control lens distortion here as well. Free Transform is a special case already, so it doesn’t seem like it would do harm to move this out of a single layer modal filter dialog.
All the PanoTools derivative programs work great if you’re rotating your camera around or close to the nodal point. This is also where Auto-align layers works best. But people don’t always take pictures this way and often for good reason; parallax can be used in many interesting ways despite being more often thought of as an error.
Great work on the SO and especially the new smart filters. The insight into the decision process for all this is helpful and interesting!
[Cool; thanks for checking it out. --J.]

By: dd

dd — Fri, 29 Dec 2006 15:02:22 +0000

thanks, John, for details and serious take on non-destructive workflow in photoshop. it’s a hard and very important topic in my workflow :)
yes- lots of small nuances has to be polished to make general experience even better than it currently is in photoshop.. i really like how adobe is improving photoshop when adding serious new functionally (like smart objects) as ‘another layer’ on top of previous functionality and the way it works – if you want, you can improve your workflow and start using additional features or you can just ignore them and new things wouldn’t force you to change anything.. photoshop is too widely used application to suddenly flip core principles and all revolutionary addons has to be carefully integrated.
although sometimes i scratch my head why _adding_ completely new features which wouldnt’ collide with older photoshop workflow – ends with the result as plugin/filter finally. e.g. ‘vanishing point’, ‘lens correct’ from cs2 or even older ‘extract’: i hate how you have to work in new window with separate controls and feel, apply your changes, get results, have no undo (e.g. either post or after result) nor ability to fix/change some steps, nor be able to somehow save operations at least to reapply them later or on another layer. imho- ‘lens correct’ transform functionality had to be added in to edit>transform commands similar to ‘free transform’ where you can change values in option bar, run on multiple layers, etc. ‘vanishing point’- in my dreams would be a _regular_ group of tools in toolbar, adding planes as ‘plane layer’ to the main layer palette, be able to save them into normal psd/psb file and as for the usage- combining new cs3 cloning improvements with an additional option to either clone normally or use some ‘plane layer’.
as i’m on the way to spill over what weighs the way i work, let me add:
‘lens correct’ behaves differently than any other filter and i think it’s a bug: after applying it, you wouldn’t find it as filter>last filter (neither last filter shortcut work) nor starting it with ‘alt’ pressed will load parameters from last apply – i’d use this frequently and always step on this bug :(
free transform’ warp mode: that 3×3 grid limitation bothers me – couldn’t we get an option to freely choose from 2 to 9 at least divisions per axis? also point/tangent translations works different to normal bezier curve implementation like in illustrator or photoshop’ pen tool: it needs manual tweaking by hand of each point since every point/tangent is treated as a very individual, e.g. clicking/moving a point and having tangents offset relatively- isn’t possible. make an option/shortcut at least. ability to convert a point into smooth/corner is also a must.. so in general- ‘envelope’ functionality in illustrator is what we need as a real warp solution in photoshop.
another thing with distort tools i’d find useful: there are plenty of times when ‘free transform’ with initial bounding box in square isn’t convenient/flexible enough. e.g. you select one side of a box in perspective and want to precisely rotate it some more – once launched free transform appears as a planar quad while i’d be glad if i could have _initial/undistorted_ square with corner points closer to selection’s in perspective.. this isn’t any kind of magic: just add an ability to hold some shortcut and be able to _transform_ ‘free translate’ corners _without distorting_. releasing the shortcut and later moving points would distort selection based on new initial form. ..also if initial grid shape could be also changed for warp mode or it’d be possible to use ‘plane layers’ from my previous dream as a start shape for free transforms/distortions of current selection- i’d kill for such abilities :)
one more: crop tool and ‘perspective’ option which inverts-perspective transforms :) it’s great but.. how many times after seeing the result i wanted to go back and just tweak a little bit some point but had to start over since step back is document before crop operation and getting crop shape form back isn’t possible.. could we get a..preview at least how the result is going to look if intelligent stepping-back is quite complicated? e.g. just another checkmark like ‘shield’- if enabled photoshop would show how the result will look if applied.
now i’m sorry for such off-topic beginning but all this was about photoshop improvements and even making it more powerful without changing core ideas or replacing older heritage or making it non-simpler by the look..
closer to non-destructive issues:
another limitation/bug regarding smart objects i find a pain (the first is- ability to link to mask :) – somehow ‘free transform’ while allows to scale/rotate or even warp smart object – fails to provide perspective distortions. free distort of corners shortcuts only skews :( the partial solution is to use warp mode to move corners freely but than these damn tangent points repositioning are required..
could we get linked masks and perspective distorts in smart object in cs3? pleease ;)
as for the future of smart objects and paint on before-or-after distortions: personally since i understand general smart object concept- i don’t find it too limited when i can’t paint directly or edits are done in new window (although being able to edit similar to ‘edit in place’ as in flash, even if only scale/rotate/transform operations would be still applied while perspective distorts, filters and anything else get ignored/disabled – would be enough for 90% of situations).
to finish that “90% enough” version: if smart filters and adjustment layers could be applied as stacks to layers (which could be regular or smart object) or layer groups (it could probably internally flatten all layers inside before processing adjustments/filters stack) so we could for instance: add blur -> adjust alpha/transparency levels -> add emboss non-destructively (and with separate masks for each layer/filter/adjustment ..i said _masks_ since one mask per layer sometimes is too less and as a workaround- i just group layer and add another mask for the group. ability to set mask operation like in after effects – add/subtract/difference/etc would be also great).
i wouldn’t mind if when i choose non-destructive way of working and add filters/adjustments in realtime – it’d take more resources and/or i’d have to wait sometimes to proceed further. e.g.- for those who would be working in old-way – nothing would be different: you apply filter, it calculates and modifies layer. however if you add filter to a stack (it doesn’t matter is it filter or color adjustment) – photoshop would calculate the result and all the time internally would store native/before any processing layer data and the final with all stack manipulations applied (it’s not necessary to store before/after between each operation. there could be even option in preferences- you either save full data with your document which stores all original “before” layers data and final stack results of each layer – so you just load and get everything. or you save only original “before” data, have smaller file, but once opened photoshop would “render” stacked processes for all layers and create cache for further modifications). if user decides that it needs to paint/edit text/etc e.g. change original layer data: stack filters/adjustments would get temporally disabled and re-calculated again after changes (the more operations you add- the more you’d have to wait until photoshop processes all operations from the updated “real” contents of layer). changing some filter parameters in middle of stack would work similar: photoshop would just calculate all operations from original layer data to show the result. there could be an new option similar to ‘preview’ checkmark which exist in every filter- that would control if user wants to see the final result of complete stack processed or only up to current operation. for example on that blur>levels>emboss stack: while changing levels it’d be useful sometimes to see how final emboss will look or only blur>levels result.
actually all these ideas are very close to how after effects with layers and filters work (they should add
ability to rasterize result to layer data, all those pixel pushing photoshop tools, selections, real free-transform, cmyk mode and could call it new decade “after photoshop v11″ :)
so basically you just have layer data and (re)process it if/when needed. the question is only flow priorities: masks>filters>transforms or transforms>filters>masks which makes huge difference and such control is a main thing i miss in after effects compared to real node-based compositing applications.
there’s a short reply to yesterday’s comment:
Mr. Adobe: have a look at Shake, and learn on how to manipulate pixels and colors.[Does Shake let you paint directly onto filtered pixels? I don't know of any apps that do, but I'd be happy to look at some examples. --J.]
in general- yes, you can add ‘paint’ node and have abilities to paint, clone, erase, etc. the result calculated up-stream to paint operation. John, if you’ve never seen in details real node workflow- grab a shake trial from apple.com and check yourself. the general concept of node trees is very easy to grasp – you wouldn’t need to learn what particular node/operation does or build complex trees to make hollywood magic, but you should feel the power and unrivaled flexibility in such way of working. i’m not pushing that photoshop has to be this (..better after effects would be :) but it’s a very strong example in pure non-destructive image manipulations..
wheew.. sorry for such a long post :o

By: Sao_Bento

Sao_Bento — Fri, 29 Dec 2006 12:49:55 +0000

After Effects is the model. Make it work like After Effects. Smartly cache objects when the pixels are directly edited like AE does with text layers that are converted to outlines. [Ah, but does AE let you paint right onto transformed media (e.g. erase a hole in a solid that's warped, rotated, etc.)? I don't think so, though I admit my AE skills have grown rusty. Letting users directly edit pixels at the output end of the pipeline is a dicey prospect, and apps (like AE) generally require you to edit the untransformed pixels, then look at the transformed result. So, the upshot is that I think PS already *does* act like AE. But AE users are already accustomed to ways of working (nesting & so forth) that have been unfamiliar to most PS users. That's why their addition to PS may feel odd. --J.]

By: darkjedi235

darkjedi235 — Fri, 29 Dec 2006 10:19:28 +0000

I have to take issue with the false dichotomy between simplicity and power. Einstein’s famous formula is incredibly simple but quite powerful at elucidating deep concepts. Google’s search UI is flat simple but shields a powerful algorithm and background complexity. Heck, a paper clip is alarmingly simple but nifty, clever and powerful little gadget. As are chopsticks.
So why not Photoshop? Through thoughtful reduction of core features, modular (community-oriented) components that extend more features, elegant UI design hiding all the exposed complexity, PS could become very simple but still retain the power.
Just do less better. And allow for more to be done equally well.

By: Rod Wynne-Powell

Rod Wynne-Powell — Fri, 29 Dec 2006 08:56:20 +0000

Hi John,
Earlier in the year I had a long spell working with some photographers in southwest London on what turned out to be three jewellery (jewelry for US readers!) catalogues for Goldsmiths.
I was mainly concerned with the watches, for which Smart Objects was absolutely the way to tackle the job.
A page might consist of from three to eight different watches from the same stable. Each watch would have been photographed several times with differing lighting and/or exposure, but without moving the watch. Invariably the winder would have been pulled out to ensure the hands remained in the same place (all the hands would be set to give the best angle – eight past ten). They were supported on perspex cylinders, and shot against white, with reflections.
The various images would then be layered into a composite file.
Each watch composite would have a series of masks to cater for the overall outline, the main face, the bezel, (often smooth and rounded, possibly inset with jewels, which were lit with sparkle), the strap the winder, the hands sometimes, and the watch ‘created’ and retouched. this would then be grouped as cutouts ‘watch and reflection’.
A new canvas for the final group would be created at the final size and resolution, and each watch would be brought in with its reflection, and immediately made into a Smart Object. Each SO would be named (or rather reference-numbered) and placed in position according to the layout, the final position was approved by the photographer, and the reflections would be 100% desaturated, and foreshortened with a grad mask. If a watch had to be moved to look as if it overlapped then additional copies of elements were added, so that a watch might seem to be behind the face of another, yet in front of its strap.
Any balancing of colour or alterations of contrast, all these could be carried out losslessly using the Smart Objects. A JPEG was submitted to the client for comments and alterations, and any changes were carried out with ease, before final flattening.
In this way every page of watches had each item at its best without sacrifice. It was a great way to work. The only snag was when an SO had a mask in the final comp, I had to move it using Free Transform and then carry out the ‘Transform again’ on the mask.
So thank you very much for Smart Objects!
[Cool--thanks for the details, Rod. --J.]

By: Tom

Tom — Fri, 29 Dec 2006 08:15:39 +0000

I may be one of the few people who did but I actually used some of the web features in Image ready for quick website creation (I am not a web guy) since I was very familiar with the photoshop interface. One of the drags was that it did not support SO. Since it looks like Fireworks will replace imageready is it likly to support directly opening a PSD with a SO.
Thanks and keep up the blogging it is great info.