YOUR ACCOUNT

Messages 1 - 45 of 57
First | Prev. | 1 2 | Next | Last 
Login or Register to post new topics or replies
voldemort
voldemort
Posts: 872
Filters: 649
It would be nice if I could set the blend mode for each layer in multi blend as is to get the effect your after you might still have to use individual blend modes maybe one layer you want screen while the third layer luminosity etc...
lets all whine for a wine port
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Yes, but the component would be huge -- we decided to keep its size manageable.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
I desperately need this, even if it only has four layers. (You can call it 'SuperBlend' and leave MultiBlend as it is.)

I'll try to explain the difficulty I'm having with the current system...

Here's Photoshop's layers palate and the result of four layers all set to "lighten". (The first layer is partially transparent.)


The resulting 80 lightness is what I expected.



Now here's a simulation of FF's blend abilities. You must in effect "double up" if you want to blend more than two layers with a blending mode. In doing so you lose transparency information which completely changes the result in some instances.



The current filter I've been working on "stones" requires a blend of four layers very similar to this example. Without the enhanced capabilities the filter will lack the depth I hope to achieve.

*Download this photoshop file if you want to experiment for yourself.
http://uberzev.googlepages.com/example.psd
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Bump!
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Vlad?
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Yes, yes. We experimented with a Multiblend Component with modes, but the results were, well, 'suboptimal'. It was just plain ugly.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
P.S. Looks like the damn forum is broken again.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
Yes, yes. We experimented with a Multiblend Component with modes, but the results were, well, 'suboptimal'. It was just plain ugly.
Ugly? As in confusing?
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
No, just plain ugly. Not the results -- I meant the component itself. And also confusing.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
No, just plain ugly. Not the results -- I meant the component itself. And also confusing.
Can't there be something like an "advanced" category for components that are confusing or too sophisitcated for the beginning user?

I'm only asking for this because there is no workaround available.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Quote
uberzev wrote:
I'm only asking for this because there is no workaround available.


Cascading Blend components?

Technically, FF uses the same Blend formulas as Photoshop does, so all you have to do is to cascade the blends in the right order. The order is crucial.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
Cascading Blend components?
That doesn't work for the 'lighten' blend mode when used with transparencies.

Scroll up on this page for an example. If that's not clear I can try to make another demonstration.

I realize this is an issue unlikely to effect most users. However it's severly dampening the potential of my Stones filter...
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Regarding your Photoshop example -- either you're measuring the resulting lightness incorrectly, or your last layer is not completely transparent. The result can't be 80% because the result of your example (both in Photoshop and FF) should be exactly the lightness of your brightest color, #3, which is, according to the color picker, 68%.

Lighten is a dumb conditional mode. It simply compares the RGB channels and chooses the brightest of the two channels for the result.

Please check your measurements or tell me the exact color and transparency of your last layer.

  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Or, which is better, attach the PSD file here -- we'll save tons of time.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
Or, which is better, attach the PSD file here -- we'll save tons of time.

Here...

http://uberzev.googlepages.com/example.psd

Merge layers 1&2 and then 3&4 and you'll note that the result is different than it was with four discreet layers.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
I'm getting 82% of lightness with my FF experiment:

Blend Test.ffxml
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Quote
Vladimir Golovin wrote:
Merge layers 1&2 and then 3&4


This is an incorrect order. See how it's done in my filter above. The correct order of merging is:

Merge 3&4.
Then merge the result with 2.
Then merge the result of the last merge with 1.

This merging in Photoshop results in exactly 80%.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Ok, that example sucked. Try this FF snippet and maybe it will make more sense to you.


example.ffxml
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Here's how it would look with a true multiblend....


Here's how it looks using the "cascade" method...
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
I can't see the second pic. Anyway, the cascade method (or the True Multiblend, this is the same thing in different disguises) works exactly the same way as does the layer blending in Photoshop.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Please check your email Vlad.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
The example you provided doesn't use the cascade method -- actually, I couldn't figure out a thing. See my filter above for an example of what I call a cascade method. Use simple examples -- they provide less room for error, both in construction and in measurement.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
The thing is what you've done is only added the alpha information at the last step. What I'm trying to get at is I will have to deal with opacity at every step and once you've blended once you've already lost some of it. \

I apologize that I'm doing a terrible job explaining this. smile:(
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
What I'm doing is "embedding" saturation data into a set a greyscale images. The problem is that it only lasts for "one step". By doubling up the blends the very subtle data is lost.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
I noticed that you use Levels with very, very close black and white points -- this pushes the algorithm too close to the FPU precision boundaries, so you may get all kinds of weird stuff with this setup.

Again, blending has nothing to do with the errors you're getting.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:Again, blending has nothing to do with the errors you're getting.
I disagree. I've reproduced my snippet in Photoshop and it shows how a true multiblend is the only way to achieve the effect I'm after on a consistant basis.

This example shows four identical circles with very faint shadows on four seperate layers all set to darken. The only difference between them is slightly different lightness values and offsets.



Now to process the image: Flatten Image > New Layer > Fill layer with Red > Set Mode to Luminosity > Flatten Image > Equalize...



There is simply no way to consitently get this effect if you have to use more than one blending step to merge the four layers.

http://uberzev.googlepages.com/Example2.psd
  Details E-Mail
Crapadilla
lvl 52 Filter Weaver and Official "Filter Forge Seer"

Posts: 4365
Filters: 65
--- Crapadilla says: "Damn you, stupid redundant feature requests!" ;)
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Zev, I think I finally figured out what you're trying to do (my today's working day is already 30 hours and still counting, so intellectually I'm not very capable at the moment -- sorry for being so dumb before smile:D ). Yes, it looks like there's indeed a problem, but I'm still unable to localize it. Most likely, it has to do with precision loss somewhere.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
Yes, it looks like there's indeed a problem, but I'm still unable to localize it. Most likely, it has to do with precision loss somewhere.
Yes there is a loss of precision (saturation data). Its inherent when using more than one step to blend multiple layers in this manner.

This problem is not "bug", as the result is expected given the situation.

IMO the best way to understand and test this is to actually code a true multiblend for FF and see if the workarounds accurately duplicate its functionality.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Quote
uberzev wrote:
as the result is expected given the situation.


This result is not expected.

Quote
uberzev wrote:
the best way to understand and test this is to actually code a true multiblend for FF


No need. You can 'code' the true multiblend right in FF by cascading the Blend components in a correct order. Photoshop does the same -- it has a blending function that blends TWO layers. It uses this function to blend ANY number of layers. It doesn't compile the blending routine for the entire layer stack into actual hardware CPU instructions every time a user moves a layer -- that would be ridiculous.

I'd suggest to split this problem into two problems: precision and blending.

Let's set precision aside for a while and test the blending first. Can you provide me with layered blending examples that don't get close to precision boundaries which cannot be reproduced in FF?
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
Can you provide me with layered blending examples that don't get close to precision boundaries which cannot be reproduced in FF?
Yes I can. The result would be correct if there was a true multiblend for the four offsets...



example3.ffxml
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Yes, this example is better. I can see the flaw even without applying Levels.
BTW, have you tried the same thing in Photoshop with a 16-bit image? (It's a shame that their 32-bit mode doesn't support layers).
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
Yes, this example is better. I can see the flaw even without applying Levels.
BTW, have you tried the same thing in Photoshop with a 16-bit image? (It's a shame that their 32-bit mode doesn't support layers).
Posted: February
Yes, it works fine in Photoshop as long as the four layers are blended in one step.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Zev, the flaw is in your method, not in FF.
Try the same thing on a 16-bit image in PS.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
Zev, the flaw is in your method, not in FF.
Try the same thing on a 16-bit image in PS.
I already have tested it in Photoshop and it works. Scroll to the top of this page for an example. (Yeah I switched from Lighten to Darken, seems to matter for Photoshop but not in FF)

Anyway like I said before if you make a proper multiblend in FF you'll get the right results.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
I see the flaw in pure PS, on a single blend with two correct sources, on a simple 8-bit image. The .psd is too big to attach here -- I sent the file to you by email. The flaws are visible to a naked eye, no Levels needed:

  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Vlad, while that might be a 'single blend' its not a, 'all at once' blend which is what this effect requires. You've broken it into two seperate merge steps there.

I'm trying to explain that the shadow data only lasts for one step. By 'cascading' or 'doubling up' or using any other alternate method to a 'all at once' blend there will be a loss of data.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Man, I give up.
There is an obvious flaw in the method.
The PSD file I sent you demonstrates that clearly.
I have nothing to say about this anymore.
I can't argue with people who can't see facts.

Sorry for being blunt.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Vlad, the method does have a flaw. You can duplicate the problem in Photoshop (by blending in 'two seperate steps' ala FF) which proves this fact.

This is why a new method is required, one that uses a true multiblend component. I've already demonstrated that this new method works correctly in Photoshop. Now I just need access to that capability in FF.

*I can see facts, I know this is frustrating. Trust me I might sound thick headed but I know what I'm talking about.
  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Anyone else want to try and wrap there head around this? Sometimes another perspective can shed light on an issue.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
I took another look at your example -- rearranged the blending order into a proper cascade. Looks much better:



  Details E-Mail
uberzev
not lyftzev

Posts: 1890
Filters: 36
Quote
Vladimir Golovin wrote:
I took another look at your example -- rearranged the blending order into a proper cascade. Looks much better:
Yes, if you know ahead of time the proper order its easy to manually set it correctly.

The reason I need this to work automatically is because I'm using a bunch of randomly colored shapes. There is no way for me to know the correct order and even if I did one adjustment of the variations slider would throw everything off.

I hate to sound like a broken record but coding the true MB in FF would be a usefull way to see if that functionality was truly redundant like you claim it is.
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
The picture above proves the fact that FF blending is CORRECT.

However, it also proves the fact that having everything rendered in double-precision floating point is bad for this particular effect.

When you do 'one-time' blends in Photoshop, it kills subtle changes in color due to its 8-bit precision.

When you do the same in FF, all these subtle changes are preserved, and when you use Levels, they ruin the effect you're trying to acheive:

  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Here's your modified example with a proper blending cascade:

example3-proper.ffxml
  Details E-Mail
Vladimir Golovin
Administrator
Posts: 3446
Filters: 55
Quote
Vladimir Golovin wrote:
I hate to sound like a broken record but coding the true MB in FF would be a usefull way to see if that functionality was truly redundant like you claim it is.


No it wont.
Reducing the precision from floating-point to 8-bit per channel would be.
  Details E-Mail

Messages 1 - 45 of 57
First | Prev. | 1 2 | Next | Last 

Join Our Community!

Filter Forge has a thriving, vibrant, knowledgeable user community. Feel free to join us and have fun!

33,711 Registered Users
+18 new in 30 days!

153,531 Posts
+36 new in 30 days!

15,347 Topics
+72 new in year!

Create an Account

Online Users Last minute:

28 unregistered users.