CS3 - Smart filters not so smart???
kwalsh
Registered Users Posts: 223 Major grins
So I've been playing with CS3 beta and it has some cool features for sure. One I was really looking forward to was the Smart Filters option in which you can use filters as a smart object. The intial thought (as I've seen others indicate) is hey - sharpening is like an adjustment layer now. However, in the end this seems to not really be true and the whole feature just doesn't seem to add anything to or integrate well with my workflow. Here are some of my problems, perhaps others have worked around these issues?
1. The single mask. Ugghhhh... This really blows. Frequently I mask high amount low radius USM to not enhance noise in OOF backgrounds but do a low amount high radius for local contrast enhancement over the entire image. Not really possible with this. Would have to have separate layers to do this which just makes problems 2 and 3 below worse.
2. Wrong order of operations. Currently the issue I have with my current non-destructive technique (which is at the end merge all layers into a new layer, apply sharpening and then luminosity blend) is that you have to do the sharpening layer over again if you adjust anything below it (if you don't the luminosity blend of the "old" layer will mix with the new changes below muting the change). Unfortunately as far as I can tell it is the same story here. If I make the background layer a smart object then I'm applying sharpening to the base layer which is a no-no (i.e. all the adjustment layers above will enhance/molest/create sharpening halos). So I'm back where I started, making a merged layer at the top of the layer stack, making it a smart object and then sharpening. Cool, I can change the sharpening amount if I want, but I still have to recreate the layer if I change anything below it. So it really doesn't act like an "adjustment" layer at all. It just lets you redefine changes to that particular smart layer, which is somewhat useful but nowhere near as useful as a true adjustment layer.
3. Holy giant files, Batman!!! Whoa, this one blew my socks off. If I have a 8MP 16-bit file set to NOT maximize compatibility (i.e. please don't save a flattened version of my file as well) I get a flie size of about 40MB. Convert it to a smart object and apply sharpening and it goes to ~115MB. What the heck is going on there???? There isn't actually any new image data around, still just the background layer. In fact, two layers in a non-smart file would still be a lot smaller!!!! This is the least smart thing about the smart objects I've found.
So, anybody else played with these things and figured out how they might help their workflow? I'm not saying they are useless by any strech, just for me they seem to be a very minor improvement with a huge file size penalty.
Something I missed? Someone else got a cool workflow?
Ken
1. The single mask. Ugghhhh... This really blows. Frequently I mask high amount low radius USM to not enhance noise in OOF backgrounds but do a low amount high radius for local contrast enhancement over the entire image. Not really possible with this. Would have to have separate layers to do this which just makes problems 2 and 3 below worse.
2. Wrong order of operations. Currently the issue I have with my current non-destructive technique (which is at the end merge all layers into a new layer, apply sharpening and then luminosity blend) is that you have to do the sharpening layer over again if you adjust anything below it (if you don't the luminosity blend of the "old" layer will mix with the new changes below muting the change). Unfortunately as far as I can tell it is the same story here. If I make the background layer a smart object then I'm applying sharpening to the base layer which is a no-no (i.e. all the adjustment layers above will enhance/molest/create sharpening halos). So I'm back where I started, making a merged layer at the top of the layer stack, making it a smart object and then sharpening. Cool, I can change the sharpening amount if I want, but I still have to recreate the layer if I change anything below it. So it really doesn't act like an "adjustment" layer at all. It just lets you redefine changes to that particular smart layer, which is somewhat useful but nowhere near as useful as a true adjustment layer.
3. Holy giant files, Batman!!! Whoa, this one blew my socks off. If I have a 8MP 16-bit file set to NOT maximize compatibility (i.e. please don't save a flattened version of my file as well) I get a flie size of about 40MB. Convert it to a smart object and apply sharpening and it goes to ~115MB. What the heck is going on there???? There isn't actually any new image data around, still just the background layer. In fact, two layers in a non-smart file would still be a lot smaller!!!! This is the least smart thing about the smart objects I've found.
So, anybody else played with these things and figured out how they might help their workflow? I'm not saying they are useless by any strech, just for me they seem to be a very minor improvement with a huge file size penalty.
Something I missed? Someone else got a cool workflow?
Ken
0
Comments
the cost is higher when you talk about any filter that involves a radius, like sharpening or shadow-highlight. Normal adjustment layers only compare one pixel to the same pixel on all affected layers. A filter like sharpening compares not just a one-pixel stack across all layers, but all pixels within a radius on all affected layers. That is a lot more than a one-pixel stack, and is an incredible amount of additional processing to do compared to a standard adjustment layer like Curves.
I've come to the conclusion that I like smart filters a lot. I am glad to have the flexibility...when it's necessary. But as a file becomes finalized, I might start merging layers and filters down if I am uncomfortable with the file size. And on some not so important jobs, I might not use smart filters at all.
I'm not sure the file size is related to cacheing, though it is a good thought. The file grows huge even if you just convert to a smart object without actually applying any filters or changes as all. Same exact pixels, just three times as big!!! Must be some sort of overhead associated with the smart object I don't understand.
Anyway, yes I think they are useful still - particularly while actually working on the image having Shadow/Highlight as a smart object is good. I'm just having trouble thinking they are da bomb as far as most of my workflow goes.
Ken
Have to agree with everything you've pointed out about the "not so" Smart Filter, Ken.
It also seems like code-bloating to NOT integrate this into the already-existing Smart Objects feature. At the very least it's redundant and at the worse, convoluted. Why keep these features separate? If you change a layer to a Smart Filter, it will act identically to a Smart Object! What possible advantage is there in this? Also, each filter that is added to a Smart Filter "group" defeintely NEEDS it's own mask.
It was very disappointing to discover that you couldn't have a Sharpen filter combined with a Gaussian Blur and mask one from the other. There's only one mask that only affects the layers below them. Ugh!
Also, you can add ANOTHER mask to the Smart Filter layer itself and it will also only affect the layers below it. What's the point of it automatically adding a mask to the filters and also allowing you to add one to the layer... that does exactly the same thing?! Totally unnecessary redundancy!!! The ONLY way it would make sense is if a mask was automatically added for each filter that was added to the group AND affected the filters below it. Otherwise, why bother?
It feels like this isn't the kind of quality we usually see from Adobe. Someone dropped the ball. It's a great idea but so far seems "half-baked."
The current smart filter entry seems to confirm how I feel about smart filters: While it's obvious to criticize how limited they are, we're sort of lucky to have them at all.
I think you all are underestimating the technical difficulty in making smart filters at all. To get useful performance from this type of feature, is really, really hard. I've read rumors that Adobe has been working on the technology behind this for many years and only now has it reached a point where we can start to use it for some things.
It is highly likely that the first release we get of smart filters will be a series of tradeoffs that were required to make them work on today's hardware. They will be extremely useful for some things, but not a 100% replacement for the way we used to do things. Over time, hardware will advance and algorithms will get even smarter and smart filters will probably eventually be a complete replacement for regular filters. But, I wouldn't be surprised if that progression takes 2-3 releases (e.g. 3-5 years). So, rather than moaning because it isn't every single thing it could be, we should start understanding what circumstances it would be really useful and be glad that Adobe is moving in this direction.
For more info on this, see John Nack's blog on smart filters (product manager at Adobe). And see this Russell Brown movie about them.
Homepage • Popular
JFriend's javascript customizations • Secrets for getting fast answers on Dgrin
Always include a link to your site when posting a question
Yeah, I don't really expect them to get this entirely "right" on the first pass, there are many technical issues to be sure. The blog pointed out a lot of UI issues I hadn't thought of since I only think of photoshop in photography terms. The design team is targeting a whole host of different uses.
I guess the long term issue is that for my uses I don't think filters really belong as a "smart object". Smart objects allow for vector based rescaling and hence all the UI and technical issues pointed out. Since almost all filtering boils down to convolution (a much simpler problem, both techincally and from a UI perspective) forcing the filters into the smart objects makes them carry around a whole lot of extra baggage that really doesn't apply to them. From one point of view I understand wanting to unify the concept with smart objects. At first blush that seems like a poor fit - but they've looked at it a lot longer and it sounds like they already tried fitting it into something like adjustment layers without much success.
Well, I'll use it for shadow and highlights until I'm satisfied and then convert it to a regular layer. Too bad it doesn't appear to be very useful for much else. Then again, give it time I'm sure people will find something it is good for...
Ken