[Steve Bentley]

You can also do a multi pass render with motion blur. It doesn’t attach the blur to the pixels but instead renders out a vector pass that you can have after effects figure out the blur. Its best to render out in a high color depth format like EXR.
This lets you render with the standard renderer nice and quick and then you can control the MB after the fact.

It takes a bit of explaining so heres a link to a tutorial https://greyscalegorilla.com/tutorials/your-depth-pass-is-wrong/
You can ignore all the stuff about depth passes.

The key items are:
Use two copies of the file in your project window in after effects (assuming you packed the multi pass into one file on render).
One is used for the color pass and the other is used for the motion blur pass (you must “interpret footage” this MB copy as “preserve RGB” in the color management tab to keep all its 16 bit goodness).
The depth of the file (EXR set to 16 bit in blocks of 16 scan lines).
Separate the layers of the motion layer with Extractor which comes with AE (under 3D effects).
Then use an adjustment layer to set the motion blur with something like Real Smart Motion Blur (RSMB) or CCvector blur
The stack looks like this:

Layer1 -(eyeball off) the Motion blur “channel” of the multipass render with Extractor set to motionvector.Red motionvector.green and motionvector.blue in the appropriate channels
Layer2 – (eyeball on) Adjustment layer with RSMB on it -the default is fine but the motion layer should be set to the above layer
Layer3 – (eyeball on) the RGBA layer of the multi pass render

Pre comp all of this and use the pre comp so the RSMB only acts on that bottom clip.
It seems messy but once you have done it a few times you’ll wonder what you did before.

[Steve Bentley]

Keep in mind we weren’t too thrilled with the vector blur in R15. It was fine as long as the object was always in motion but once it slowed right down or stopped there was always this little bit of blur on it. Or it popped from the frame before dead stop to the dead stop frame as that little bit of blur suddenly vanished.
If you have the time, using the physcial renderer with the motion blur turned on did a better job – but it does take longer. (you will only see max two buckets working as it renders no matter how many cores you have)

[Steve Bentley]

Keep in mind too that the metablobs don’t have to be crazy smooth as a mesh to look smooth in the render. Just adjust a phong tag to a higher or lower number to get the smoothness you are after. With the correct angle setting even a cube can approximate a smooth sphere.

[Steve Bentley]

Do you have the Mograph module? It works well with hair (In fact, if your not doing actual coiffed hair we prefer the mograph module for controlling the hair particle system over the hair system itself). You can have your money object appear anywhere along a specific “branch” of your hair splines and using a “random” or “shader” effector have that money appear at random times, sizes, orientations and even with different deformations (so some are curled and some are not).
You can also use the “wiping” feature in the effectors to wipe the money on from lower branches first to upper branches last.

[Steve Bentley]

And I take it you don’t want the letters to bounce all glued together?

Often for this type of thing, where accuracy matters at the beginning or end of the dynamics sim, we’ll bounce something else. Have an invisible stunt double for each letter do the bouncing instead of the visible letters themselves and then, through expresso, link the star letters to the stunt double’s motion data and only access the vectors you want. You can also use a range mapper node to ease in and out of the magnitude of what the stunt doubles are doing (so no affect at the beginning and end but follow lock step in the middle of the animation). At the end of the animation position your letters where you want them and then link their nodes to the stunt double’s. The visible letters will follow the moving doubles all the way back to the beginning and you can massage how they move. Realistic physics provided by the sim and accuracy provided by your keyframes.

If you want to take it one step further you can nest the stunt doubles to a parent that is animated and the parent kind of steers the stunt double around so that you are adding in some x or z translation on top of the dynamics data to make sure they end up in the right spot. This way you get the best of both worlds: dynamics and trained performing letters.

[Steve Bentley]

Do you have to use shapes? If they are all the same shape or a few base shapes can you render out an element with an alpha and bring that back in to use for the array population?
I wasn’t aware of a limit with shapes – but we’ve made things like this with png’s or pshop files where we’ve had objects in the millions with no crashing issues (but to be fair we have gobs of ram).
Element3D renders instances so it shouldnt kill your ram.
Cloners and Effectors (yes that’s its name) also renders instances so it shouldn’t kill your ram either but I’d have to test it.

[Steve Bentley]

I too am a little foggy on what you are after but here’s a few thoughts that might help.

I don’t think the expressions are going to pick up any shift in the pixels that an effect like corner pin will have on them. Applying the effect doesn’t change where the corner pinned objects anchor is, and depending on how that corner pin was created or how extreme the warp, those pixels can be a mile from where they were originally but the anchor point hasn’t’ shifted at all.

Conversely getting the position of a dot and putting a corner pinned object at that spot may not align the way you might think since all the pixels of the effected subcomp have been skewed and no longer match its original anchor point – which is what the expression is really interested in to match the dot position (but if they are all warped the same you could build in an offset to compensate).

If you can cheat, the transform effect (in distortion) might keep that shifted position data as it processes its skew a little differently than corner pin.

Lastly if you can help it don’t use the corner pin effect. Its quite soft. I don’t think its bicubic math is near as good/sharp as the CCpower pin the comes with AE.

[Steve Bentley]

You can also get there with Particular. But with Pastiche or Particular or any 2D (2.5D) plug in you will have an issue. If it’s the classic SIM look of isometric people you are after and they have a walk cycle associated with them, the perspective will only work if they walk on that angle for their entire “life” as they journey towards or away from their target point on the logo. For most particle systems that’s going to be tough to keep that vector of movement consistent and correct for each persons correct angle. (you might need a different particle system for each specific angle of a set grouping of instances of people – but then you are going to have overlapping collisions issues.)
Element 3D could do this too and you would have the advantage of as-near-to-true-3D as you can get in AE but you would have to have animated 3D people to start with. But the 3D aspect of Element would get rid of the need for perfect vectors for the people.
How we normally do this is in a 3D package with a plugin/program like Massive that is built for crowd simulation. C4D’s mograph could also do this with an animated set of people to clone. The beauty of Massive is that the clones are kind of built in.

[Steve Bentley]

First I’m curious about the need to work in RAW format. Can you elaborate?
I’m pretty sure that once you hit the zero key and the images cache for playback, you are no longer seeing a RAW set of pixels (even if you could see a raw set of pixels on an 8 or 10 bit screen).
Would it not be better to convert the RAW into a format that is more inline with what your output codec/colorspace is going to be?
If its a fidelity issue, have you looked at either the Acadamy ACES format or the ILM EXR format? Either of these will bring down the on-the-fly conversion issues you are seeing and reduce the processor overhead significantly.

[Steve Bentley]

Re: Competitors to AE – Um, there is and have been for quite a while – although not at the same price point – Fusion, Nuke, Flame, Inferno, Shake, Natron, Combustion, Mistika; take your pick.