View Full Version : earth zoom with shake

09 September 2007, 02:47 PM
i saw this tut ( but i tried to replicate what he was doin in shake but the fisrt clip jus get blured out.

this is what i did if can understand

1. im01 + move2d over im02 then resized im01 to match im02
2. over node with im01 + imo2 now i add a move2d an resize an soforth

when i get to im05 i loose qulity all the way down to im02 an 1 any work arounds to the what andrew is doin it on this site?

09 September 2007, 09:04 PM
The trick is really about concatenation. If you look at the nodes in the transform panel, some have a small letter c in the corner. This means that Shake will compute multiple transforms before generating an image.

So what you need to make sure of is that the main zoom node is directly linked to each of the individual scale/transform nodes that you use for alignment. However, you need to have a composite node somewhere but putting it in between breaks the transform concatenation, which is why you are seeing image quality loss.

So, I would say the easiest way to set this up is to use a multi-layer node (remember this works the opposite way round from the over node in that backgrounds come first). I tend to use Move3D and multi-layer nodes all the time because you don't lose any processing speed and the extra functionality is there when you need it. You can use over nodes if you prefer though.

So yeah, attach image 1 and two to the multi-layer node and scale the closest one (e.g. street-level) down using a transform node. Now copy this transform and instead of pasting, right-click and choose edit > paste linked - in the image, this is called Move3D2_clone1. Then insert it onto the second image by holding the n key. You will see this has scaled down the further out zoom too. We need to scale this back up.

So add another Move3D node above it (Move3D1 in the picture). It's important that it's above because you need the center point of the common scale node to be in the same place - this can be confusing relative to a layer based app or 3D app because they work from top to bottom, parent to child. In Shake it helps to think of the source nodes as the children. So now on node Move3D1, scale the further out zoom image back up. notice that although you are scaling above 1, it won't lose quality. Get image 1 and 2 in alignment and repeat for all your images. Your tree should look something like this:

The pink circled node Move3d2 is the node you scale from 1 to 0.003 or whatever, since they are linked, you just need to animate the original one. I actually found this process out the hard way because I tried to do the same for a project a while back and I seriously wish I'd had that tutorial. Even a quad G5 with 4GB Ram chokes if you try to scale a 2000x2000 res picture up by more than about 5 or so. It just kept crashing the render.

Just hit i to ignore the images you are not aligning but be careful when adjusting the upper Move 3D nodes because they are not all parented like in the AE tutorial so you basically have to align 2 to 1 and then scale the main node down and then align 3 to 2 but do not adjust 2 again and so on. I guess you could have multiple linked nodes but it gets pretty messy. If there is a better way to do it, someone else is welcome to chip in but this setup seems to work and the composite tree is simple so I hope it helps.

09 September 2007, 02:11 PM
If the Multiplane node worked a bit better, it might have been ideal as you should just have your inputs and a multiplane node but like I said in the thread about the tracking, it seems to mess up alignment way too easily and it only supports one level of parenting, which is just stupid really.

So, a workaround would be to use this setup:

The M and T nodes are Move3D nodes just named for convenience. the green lines show concatenation and purple ones show expression links. Basically:

T1.xPan = T2.xPan+M2.xPan, T1.yPan = T2.yPan + M2.yPan;
T2.xPan = T3.xPan +M3.xPan, T2.yPan = T3.yPan + M3.yPan;

and so on. This means adjusting M2 will move layer 1 and 2. Then you'd adjust M1 to align layer one on its own. M3 moves layers 1,2,3 etc.

10 October 2007, 02:24 PM
Slightly off-topic, but, osxrules, are you aware that plugging a matte into the side of a FileIn node is a potentially very dodgy thing to do (I believe the Shake manual specifically recommends against it)

You'd be much better off putting an Inside node below the FileIn and connecting the matte to that.....

10 October 2007, 06:13 PM
sweetness thanks for the info i see exactlly what u mean, ill play around with what u have thanks much appreceated


10 October 2007, 02:00 PM
Slightly off-topic, but, osxrules, are you aware that plugging a matte into the side of a FileIn node is a potentially very dodgy thing to do (I believe the Shake manual specifically recommends against it)

You'd be much better off putting an Inside node below the FileIn and connecting the matte to that.....

I agree with you but the manual seems to promote input masking (p902):

"Donít Mask Layer Nodes
The side-input masks for layer nodes should not be used, as they behave counter-
intuitively and will not produce the result you might expect. If you want to mask a
layering node, mask the input nodes, or use the KeyMix node."

I don't like masking input nodes because they don't have the same controls like invert matte etc. According to the manual it's the wrong thing to do but I actually mask layer nodes - I usually mask a dummy 'over' node with nothing in the foreground. I should use Inside though (I just discovered 'over' first and picked up a bad habit).

I guess what they mean by it not working as you expect is that they expect that you expect it to mask the result of the over instead of just the background, which it does.

In the above example, I just threw those in quickly to get a circular fade, each of the images would probably have separate masks too and an inside node just after the input node is good to attach the masks.

10 October 2007, 04:02 PM
Oops - yes - you're right.... My mistake.... I'd recently read the bit that you quoted....

I'm still not a fan of masking inputs, though! ;)

CGTalk Moderation
10 October 2007, 04:02 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.