[Steve Bentley]

So here’s a comp with a few tricks in it (that are especially applicable to talking heads). Note that the project is in 16 bit even though you have 8 bit footage – this lets any color warps or level changes have a little more headroom instead of clipping.

In order to get rid of the blocky noise you have to crush the matte, this leads to helmet head. So we’ve added some hair back in at the edges. Its best to animate the head outline mask to follow the talent around (Rotoscoping). We haven’t done this but you can keyframe the mask as her head will undoubtedly bob along has she talks. In this case it doesn’t have to be a closed mask. We’re just making a glowing line where the hair is at the edge (and then hiding that glowing line) and that will act as a soft rim matte to let the edge hair through. You could also do this with inner and outer feather masks if you like, but this will have fewer mask nodes to animate.

We tried degraining the footage but grain isn’t really the problem – its the H264 blocks and general camera blobby noise that are the issue.

The main key is just a simple keylight key with some tweaking, I’ve also left on (but turned off) a few of the other native effects we tried to get rid of the noisy edge just so you might play and see what you get. We were getting better results handling the color and spill suppression on another layer unencumbered by the keyer. And the keyer controls, in this case, seemed a bit more of a big hammer than doing it another way. Just because you are keying a layer doesn’t mean you have to “see” that layer. If it gives you more options you can pass another version of that layer through the key – it’s just an alpha at that point. (you can even use a keyed layer as the alpha and then pass a copy of that layer with its own key through the original alpha – sort of applying they key twice.) Sometimes a more subtle, slightly transparent key, masked twice is better than crushing a single one too much. (but we didn’t do that here).

Ideally you should pull multiple keys on this shot for the various sections of the screen that are differently lit, and then mask off those areas and blend everything together. Then animate those masks to follow the action. Many many many green screen shots have a rotoscoped mask traveling along doing everything from filling the center of the actor in with good color corrected footage, to separating well keyed areas from areas of different color or density, to masks that remove tracking dots or wires or other appliances that were in the shot, or running a clean edge along scissoring legs that blob together as the apex closes at each stride. Unlike those behind the scenes DVD’s, it’s not just sampling the green with the eye dropper and you’re done.

Below the main key and its color element is another version of our edge hair trick – but this time with no mask. With it in behind the main alpha of the key, the wisps just poke out from behind the layer above. For both the hair edge layers: because the red record is so clean, if you shift that record to act as the alpha and then set all the others to red you get a cleaner less noisy element. (in this case we used luminance for each layer with red as the alpha just because we got a better signal to noise ratio)

Below that is the weird layer. The AltKey layer is almost the same as the top keyer, and again its turned off so we can pass another version of the main footage through. This time though we’ve blown up the main footage so that hair further in on the head (that is not interacting with the green screen) is passing through the edge of the matte. This fills in (even so subtly) hair at the edge and doesn’t have any of the green screen noise in it (or color) and again helps get rid of the harsh edge needed to clamp the noise away at the edge of the matte.

Finally on the background we’ve added some noise to distract from the very noisy front element. (we did our usual dirty quick talent collapse to get rid of her since we don’t have a clean plate)

There is one more problem you will have to animate the settings to deal with. It seems the talent is so close to the screen and the camera so close to the talent that when your crew moves about, the lighting on the screen changes. I can watch the matte get worse about 2 seconds before your clapper/loader comes into frame. In the two seconds before he appears, the screen gets darker and therefore the key changes and the settings may have to be keyframed to compensate. This darkening is him moving in for the kill and soaking up light as he comes – all things on set either reflect or soak up light. We also haven’t done any garbage mattes (for instance the bottom right corner where the key is much worse than the rest at these settings)
The final issue is a little moire pattern in her left collar (stage left). That is nothing to do with the green screen and is just an interference pattern between her blouse and the pixels in the camera.

There are about a thousand ways to do a screen shot. When we get a toughy, normally we pass it around the shop and everyone has a crack at it. We don’t share what we’ve each done with the next person so they take their own spin on it. Perhaps others will offer other solutions than we have done here (and this is by no means an extensive R&D project). In the end it has to be up to you because we won’t know whats important to you. You may be fine with a clamped edge giving her a crisp outline as long as it gets rid of the noisy edge. You may prefer a bit of noise if it gives her a softer edge and some wisps. You may prefer a blurred edge to give it a “look” and thus remove the noise altogether. And every set up you have in your shots may require a different method due to lighting, colors, talent or movement.

12293_greenscreenheadcomp2.aep.zip

[Steve Bentley]

Ok that footage is miles cleaner than the one you uploaded before. However it’s still pretty noisy.

My preference for the color of green is more than just a personal preference.
In most CCD cameras (and ALL film cameras) the blue record is the noisiest, followed by the green and then the red (hence red-screen being the best, but unusable for humans – we once did a green-skinned alien movie and we used red screens – best keys ever!!!).
If you isolate each color in your footage to have a look (use the little kalidascope icon at the bottom of the main comp window and step through each color) you can see that red is pretty smooth. green isn’t too bad either. but blue is a noisy blocky mess!. If your green screen has more blue in it (ie a darker bluer green) it will be noisier. But if you go too far the other way into the yellow greens you are getting into skin tones and blonde hair colors.

Also – was the camera locked off? There seems to be some breathing going on with the frame. You can certainly do moving camera green screen shots but with your wrinkles you are creating noise that has to be figured out in the compression, and with the slight movement the “lava lamp” effect on the matte edge will be more prominent. BTW stabilizing the shot at this point will only make that worse.
Use the heaviest/beastiest tripod/dolly you can afford.

Let me have a play with this and we’ll see what works the best.

[Steve Bentley]

You’re going to have trouble getting an unbiased answer on that topic, or, brave souls who are willing to withstand the pile-on… but consider this – 3DstudioMax, Maya, Softimage and few others are now all owned by Autodesk – (they also own Discreet Logic too – the makers of Flint Flame and Inferno). So does that mean going with one of those 3D packages will benefit from some terrific in-house trickle down? or will one or the other become obsolete or not make its quota and be consigned to history’s scrap heap? Who said: “The best way to beat the competition was to buy them and then shutter the company!”?
I’m actually shocked this hasn’t happened to any great degree in AutoDesk’s case. While there would be a mass industry-wide revolution if they shuttered Maya, I’m not sure how deep the in-roads of 3DS MAX are in the pro industry – would there even be a whimper? Perhaps that’s changed with their new owners and it’s now a going concern. I just don’t hear about MAX anymore. And Softimage hasn’t seen an update in 3 years, so perhaps it’ll be first on the block.

So there’s that- then there’s what platform you are on, what graphics cards system you have bought into and what do you want to do with your 3D package when you get it. (fx, characters, simulation stuff – water smoke fire etc, flying logos, environments, etc). Not every package works best on a particular platform, some packages work best with specific graphics engines. And certain ones are better at certain tasks and others suit generalists better.

Finally two more questions – price point and how often are you going to be head-down buried in the software?

Also consider that with some packages you pretty much have to add more sheckles for an after market renderer. The quality of render engines has exploded in the last 5 years and it’s become hard stay viable with some of the package’s built in renderers.

All of the packages have come a long way, to the point where unless you are specializing it’s hard to pick a king of the hill.
More and more are coming from FX houses who have developed their own in house tools and they sell the package (or an addon to another mainstream package) to offset development cost and seed the world with future, already trained employees. So dig a little beyond the usual suspects.

But if you are a one man show or small band of trouble makers, you might want to put ease of use at the top of your list. Some packages don’t take kindly to users with too many hats. And what’s the cost of a render farm – not the hardware but the farm licenses.

With that in mind here’s a few thoughts on the big ones: (in no particular order)

Gulp!

C4D: I find C4D crazy easy to keep up with. I can be off the box stuck in the compositing world for months and get back on the bicycle with very few wobbles. And it’s finally come of age where you can get under the hood to do some specialized stuff or link it with an outside render engine to tease out some specialized looks. But it has been late to the party on a number of fronts, so while other apps licked certain issues years and years ago and could focus on pushing the boundaries, Maxon was still playing catch-up. In the pro world C4D can get you there with most things (especially with add-on render engines or doing some tricks in 2D after the fact) but often if you want to go that extra mile, you can find yourself looking jealously over the fence at the neighbor’s shiny new 3D app. Keep this in mind too, as you are swooning over some renders you may have seen – was it a still or an animation? C4D can do mind boggling things in stills, but sometimes (and only just sometimes) when you move that remarkable thing over time, its sins become manifest.

Maya: Maya is deep – I mean Marianas Trench deep. There’s very little you can’t do in Maya providing you have a programmer or two. Not that you need a programmer or two. I’m talking about that” extra mile” stuff. With a coder at your side there are few places you couldn’t navigate. That being said, in a normal production pipeline, even without a coder, you will have few wants. And if you are using it every day you will learn to swim those depths and give Mark Phelps a run for his money. But dabbling is tougher in Maya. I know guys, who like me, have been with it since it was Alias, and they know their corner of its universe inside and out and can make that corner dance, but ask them about another area of the software and they will say without hesitation that they don’t know how to work many of the features. I have a theory, that as much as the “specialist” (as opposed to generalist) appeared in the FX industry because of the sheer amount of work, and it’s the only way to make FX production feasible/profitable, it could also be because it’s the best way to work with Maya (one of the kingpins in the FX world). In a Maya pipeline, lighting guys don’t work on animating, particle guys don’t do textures, modelers don’t write shaders (there’s another whole topic right there!). Even guys who rig and paint the character weights don’t do the animating in many shops. To get the most out of any tool I suppose you have to practice the craft every day, but with Maya I find you have to narrow your focus a little to keep the expertise up – and swim a mile down every day to stay in shape.

Softimage: This used to be the king of the Character animation apps. I’m not sure what its long term lifespan is. We used to model in Maya, animate in Softimage (and render in something else). You could do both in either but this pipeline ended up being a perfect synergy. But in the intervening years there has been so much cross pollination, everyone else caught up. I’m not sure what the user base (hence future) of Softimage is. And to be fair, I left it behind a decade ago. Again, no update since 2015. (cue scary music)

Houdini: Among the guys I know, we all think of Houdini as the TD’s (Technical Director’s) package. I always think of Houdini when there’s a problem we can’t solve with another package (usually dealing with particles or sims or environments or something we need mathematical and microscopic control over). Often you don’t even need a programmer like you might with another app. Houdini can do it all of course but it tends to specialize, even among the specialists. With Houdini you can go deep deep deep like you can with Maya but you might need to bring your PHD diploma along for the ride, or at the very least, your waterproof propeller beany. Again not a package for dabblers (or for the timid). And with a smaller user base, fewer experts to lean on when the going gets tough. But when it comes to magic – wow!

Blender: So first: Free!!!! And second – if you are diehard 3D guy you might spend your first week scratching your head. The interface is just soooo different it’s tough to figure out where stuff is. But if you can get beyond that it’s remarkable what a community effort has come up with. There are things Blender can do out of the box that rival or surpass C4D with a plug-in add-on (smoke for instance).
That being said, I always worry that in a production situation you will get backed into a corner because there is no one being paid to come to your aid. Yes the community is very responsive and helpful, and one might argue even better at dealing with user demands than some billion-dollar software company. But it can take time to get the right good Samaritan to solve a particular problem in just the right way.
On the plus side this package had tools in its first year that took Maxon a decade to implement. But I wonder, from a developer point of view: if one was developing a 3rd party render engine, would you want to put effort towards a group that expect their software for free? I worry it will lag behind in this area (or any bleeding-edge improvement in the industry) simply because it will be harder to earn development money back. But the internet is a wonderful thing and I’m always amazed at what makes money in unconventional ways.

Lightwave: I must admit I’m not current on this app. But I also don’t hear about it very much in the Biz – never have. It had it’s heyday back in Babylon 5 days and turned a lot of heads, and we gave it the ol’ college try back in the late 90’s but it was just so different from what we were used to, we chose another path. It seems to be a concept designer’s tool these days. There was always something up with Lightwave’s render engine (when it was fast we ignored this problem for the most part), and of course it could be the user too, but it had and still has a “look” (and not in a good way).

Fuze and Mixamo: I’m not sure where this pair is going but so far I’ve been unimpressed with how they looked after their user base. We used some mixamo stuff for a project because it was easy and hooked into C4D effortlessly. By the time we went back to update that project Adobe had bought it and created some vaporware called The Fuze Project, and left us all with a broken app and nebulous promises that were never kept. I’m glad to hear its being used again, but lately I’ve been helping some people who have been stung by the current package not playing well with other 3D apps (export for rendering and final animation). Given Adobe’s history of slow feature additions and fixes (no, I don’t want a darker interface in AE again this year, after two decades I want a much needed paint package!) , I personally don’t have a lot of hope for this one given that there are so many other options out there that don’t cost an arm and a leg and are more mainstream. I don’t know if putting Fuze on your resume will get you too far.

3DSMax: I can’t for the life of me figure out why you can still buy this software. Not because it’s bad or anything, its not, but when the company who bought it now owns pretty much all the other’s… It always seemed to be a hobbiests package or one that got used a lot on low FX budget TV shows. It was also probably more pirated that Pshop… I can’t figure out why they are keeping it around. I can’t believe it generates near the sales that Maya does and they must have bled it already for the tech they wanted from it. To be fair I’ve been away from Max for a long time so I can’t speak to its functionality now. I only worry about its future. When you make the leap into a package you are investing time, lots of time. So you should look to the future and pick something that will be around and/or keep you employed as a skilled user of that product for years and years to come.

Modo: Can’t comment. Seems to do some cool things but not enough for us to upend our pipeline.

Let the pile-on begin!

[Steve Bentley]

Sure upload a small chunk to drop box.
Just copying the file to a folder or uploading to the forum won’t compress it. You could just stick the original in AE and render out a small bit at full quality. If its 8 bit footage you could render out as Quicktime animation best or AVI uncompressed – both of these won’t introduce any compression (but the QT format will be smaller). If its 10bit or higher, set your AE project to 16bit and render out openEXR with floating point and RLE compression. This will render a bunch of stills.
You have to zip for drop box I think so just check your zip settings so that its just making a .zip wrapper and not actually compressing anything. The zip file should be just about the same size as the file you zipped. If its smaller you will know its been crushed. It may be a lossless crush but why take the chance when its the compression we’re investigating.
Once we’ve got the unadulterated footage i’ll send a CS6 file.

[Steve Bentley]

PolyTrans! I wonder if that can take animation data as well as the bone/mesh structure. If it can the next question is will be can it handle iClone. I know it will handle C4D native – but you might have to start in an older version then open that file in your current version.

[Steve Bentley]

Well its crazy tedious but could bring your mesh in with bones (but no animation) and then copy and paste the animation data on each bone from iClone data lists? As long as the scale remains the same, the data on one bone will jive with all the others.
MS XL can be your friend and a good pass through to massage the data to get it in a form that Cinema will understand. Before AE became C4D savvy we were throwing text based/spreadhsheet based data back and forth between the two programs all the time.
You may have to invert some of the data or scale all of the positional (but not rotational) data, but then that’s what the spread sheet massage is for. Sometimes Word can help clean up a comma or tab delimited set of data.

The problem just has to be a sort of bake on each and every vertex – its the only way I can see the Gb’s getting that high.
Is there another format transcoder that could be an intermediary between iClone and C4D? The name of the big one escapes me at the moment but if I come across it i’ll post.

[Steve Bentley]

So if a bouncing ball was in your shot you would want that ball to stay still while everything else moved? (and that’s ok, i’m just trying to determine what you are after.)

Depending on the version of AE you have there are a number of approaches but the principle is still the same. Generally this is called stabilization. Find a detail that you can “track” and then tell AE to fix that detail to one point.

Instead of sending your tracking data to the position attribute of the footage you want stabilize you send the tracking data to the anchor point (it acts as an inverse movement) So if the ball goes up in the frame, sending “up” data to the anchor point makes the footage come down thereby keeping the ball where it was. We’ll often send the data to a solid the same size as the footage and that way it’s saved and not part of the footage at all. Then we can copy and paste that data anywhere you like or make an expression that links to the data to another piece of footage or run the smoother on the data before you paste it back onto the original it came from.

You can copy position data to the anchor point and visa versa but you can’t copy rotation data to the position data or visa versa. Its all about how many elements the data has – position has x and y, so two elements, but rotation only has degrees – so one element.

I mention rotation because you can choose to turn on the rotation tracker as well so that the position stays still but so does the angle. There is no anchor point for angle in the same sense as “position”. If you choose Stabilize in the tracker, this will be managed for you. But if you just want to play with the data, you often have to do a bit of 360-rotTrackData math to get the right data to paste back onto the rotation attribute. (you have to calculate your own rotation “anchor point”.)

Be forewarned though – stabilizing can make your footage seem too small to fit your frame and you end up having to blow up the footage so you don’t see the edges. That will make the footage softer. Move footage even one pixel and you can suddenly see an edge.

[Steve Bentley]

Rereading your posts again: at that bit rate out of the camera you should be ten times better than what we saw. Grab a bit of the original camera footage and pull that into AE (ignore your edit list for moment – we just want to see if unadulterated camera originals can key ok.)
I’ll bet its the crush coming out of Premier that’s killing you.

[Steve Bentley]

Ok that’s not quite true – with the mask locked the puppet tool creates a hybrid mesh mask that incorporates the alpha (if there is one) and the mask you surround stuff with.

[Steve Bentley]

Is it possible that as you work with premier (or whatever) and AE you are rendering your edit chunks out for use in AE and therefore the footage is being compressed and recompressed and recompressed? Compressing an H264 gets exponentially worse the lower the original bit rate is. And then if you recompress that, well, its starts getting very noisy. If you must render out clips to use in AE use a truly lossless format. Since it’s coming from H264 even Quicktime Animation best is lossless and high quality for that kind of process (and its runLengthEncoded so green screen stuff has very low file sizes without loss).
Are you color correcting the files in Premier before they go to AE? You want as raw an original as you can get for the key – every time to color correct you loose information. Try it – take an image in pshop and look at the histogram – now color correct that image, flatten the correction (make it so!) and then look at the histogram again. Instead of a continuous set of curves of the colors you get a spiky skyline – all those gaps are missing data.

H264 can work (as I think that clip I uploaded shows – at a higher bit rate at least) but the idea is to get as good a record as you can of the on-set stuff because something will cause you grief and the cleaner the matte you can pull without any tricks, the more tricks you have left to fix other stuff.
Did the AE file I sent with the key in it not work out well enough?
You could try another keyer – Primatte makes a good one – not sure if there’s a demo or the cost. There’s the granddaddy of them all – Ultimatte. Sometimes another compositor (Shake, Combustion, Nuke etc) will have their own keyer that will be dialed in just right to pick out details that others miss.
There used to be a pro version of the Keylight keyer too. It could tease out that much more of your edges.

In the end though, the clip you uploaded is really noisy so there will only be so much you can do without by-hand paint repair. When it comes to fixing green screen shots, rotoscoping certain areas is not a dirty word – most of the lord of rings got roto’d even with all their blue and green screen work. And don’t forget you don’t need a perfectly opaque matte – its ok to have a bit of the white part of the matte (the head) be grey – you can always make a garbage matte that is rough and fills most of that back in and leaves a feathered edge being a little transparent, and being able to see hair wisps at the edges is far better than helmet head.