CGI Explained?

[Guest surferphil]

I'd like to explain how, what we generally call CGI is produced.

I usually baffle people when attempting a simple verbal explanations of how a Computer Generated Image is created, why, what for and what CGI actually is.

The inerrant problem with explaining an aspect of visual communication is not having a visual reference to show.

A picture paints a thousand words, but in some cases words don't mean anything if you can't see what they are talking about.

Having worked in all aspects of CGI I can explain pretty much any part of the creative industry.

Is anyone interested? Or no.

[Bibs]

That engine you showed me was mesmerising!

[slewthy]

Yes please (sitting quietly, not fidgeting and being generally well behaved)

[arbell9]

Can you explain in 3D, I have my specs on

[Bibs]

Do you wear them all the time just in case?!

[Mark Blanchard]

I use Solidworks most of the time. With Photoworks it can produce some nice renders.

[Guest surferphil]

Cool, Mark has it covered! So I'll hand over to Mark now to explain how it all works.

Only joking, I have a very brief history of where CGI came from and then I'll show an example to take apart and explain how it's made.

I'm no hack as you know, so please bear with me and feel free to send me a message if I make no sense at all.

There are many applications and industries that could be covered, but the most familiar point of view for most people; is the use of CGI in Photography. The basics for photographic quality rendering are the same for other industries and I can cover all the main disciplines and aspects of CGI (except animation) from a single image/piece of work.

I would love to dissect the Engine job I did for McLaren (Bibs referred to) but McLaren will get upset if their image is used without their permission (I won't get it) and it's not in the public domain.

I have a less exciting project I use for R and D that has no ties with the commercial world and therefore less glamorous, dynamic or corporate which is perfect for dissection This is not really about me or my work but helpful to explain the processes and techniques which is what is important for photography enthusiasts (you).

(Wish me luck)

[Vulcan Grey]

I'm interested to see. I started dabbling back in 1990 with the Amiga, using Imagine, Real 3D, Vista Pro, Aladdin4D, then later with the Video Toaster, Lightwave, then 3D Studio and Maya on the PC. Last 12 years I've just done 3D CAD and some Photoworks for product design, similar to Mark above.

[Guest surferphil]

Most know the history but for those that don't, here's a simplified version of how the CGI industry came about:

A very brief history.

CGI (Computer Generated Image) is an umbrella term for digital image creation and manipulation that really came into maturity during the 1990s, in the wake of Industrial light and magic (a Lucasfilm Ltd post production house) developing technological solutions for film effects (Termiantor2 etc), they made digital effects available to any production company who wanted their services thus becoming wide spread. Lucasfilm/ILM's digital research and development department was called Pixar.

Whilst this was going on TV, media, computer games developers and print graphics agencies were pushing the digital evolution to solve the same problems: cheaper, cleaner, reuseable assets.

As the power of computers increased the application became broader and more user friendly filtering into the main stream as costs dropped and software developed.

Today computers are used in all aspects of all industries that once used traditional analogue machines. In creative industry, computers and digital equipment are the primary tools because they are so good and cost efficient at reproducing and storing images; without degradation or loss.

Creative studios who's clients have contracted them to deliver print media, web site, film, TV Series etc hire or employ 2/3D artists to create or finish photographic quality images.

The profession.

CGI is divided into two main areas of expertise 2D and 3D, there are different jobs within these areas e.g.

1/ 2D:

Editing

Compositing

Photo retouching

Digital painting

Effects and Tracking

2/ 3D:

Modelling

Lighting

Rendering

Animation and Character animation

It's very rare a finished 3D Rendered image will not contain an element of 2D photo retouching or digital painting in the process. The industry is made up of 'specialists' who work in great detail within a small section of the industry and 'generalists' who can cover most if not all aspects of 3D and 2D work.

Get on with it!

I'll start with a still image as it's the most accessible and familiar form of CGI to everyone.

All right, this is just an example for dissection, so don't get distracted by any technical or creative opinions you may have. It's a fairly clean image in that everything is easy to see, lots of detail and light, simple focus so nothing is hidden with crappy blur or lens flares and the camera height and angle is very un biased; making this a nice simple perfect example, which is the main thing here.

Edited October 27, 2012 by surferphil

[Bibs]

Is that hand drawn, 2D or 3D? How long does it take?

[Guest surferphil]

Quite some time, it's a 3D model I built lit and rendered in Autodesk Maya. It's an on going R and D project to experiment with, in order to find solutions for my professional clients. It will never be a finished piece of work so perfect to rip apart.

So firstly I have the backplate (photo in the background) it's just one of my holiday snaps taken with a sony cybershot whilst on a surf trip in Taz/Oz. It's by no means a professional photo and sadly the camera only saves as a jpg format. It was ideal as I never used to get this sort of backplate to put a car in professionally.

This is the photo out of the camera:

Photo after a minor retouch:

I originally did a render with the puddle reflecting in the car and vice versa as a challenge and it worked well but it wasn't so nice to look at, so I removed it with clone tool in Adobe Photoshop.

You can see the shape of damp patch on the ground repeated, this is fine if in this case it will be hidden but it's always worth checking there is no visible pattern or repeat when cloning.

I rotated the image so the horizon was level and gave it a bit more contrast as it was a hazy winters (August) day and cropped.

I then copied the hills to a new layer and added a simple Gaussian blur to take the focus back away from the hills.

I'll cover the rendering and compositing last of all but before I do, here is the car composited without the backplate:

Edited October 28, 2012 by surferphil

[Guest surferphil]

Originally I modelled the car from scratch so I took some ref photos of my car getting the images as flat as possible with the least amount of lens distortion to give more accurate proportions.

I took about 100 photos of angles and details to help see the shapes when I was modelling.

This is a screen shot of Maya with the model progressing quite well.

The model is built out of polygon quads which are 4 points in an empty space that the computer can calculate as having a surface between the 4 points.

Each point is called a vertex and the mathematical plane between them is a face, ideally the plane is flat. It has no thickness so it technically only has 2 dimensions and is purely a mathematical computation. The adjacent faces share 2 vertices so the computer can calculate a smooth surface between the faces, this allows a series of connected faces to describe a curved surface; which is what we need for our car.

A 2nd mathematical calculation (mesh smoothing) can bend the strait edges so the profile of a model can appear smooth, as well as the surface. This can be calculated in real-time, which saves a huge amount of data you would otherwise need to describe curved surfaces; keeping the model quite simple in geometric terms.

To build surfaces it's easy to start with a flat plane (like a grid) and move the vertices in 3 dimensions to describe the shapes of each object.

There is nothing in a 3D scene, unlike the known universe that contains everything, so the challenge is to create from nothing and having no tactile sense or gravity to hold things together. Modelling can be extremely laborious and tedious just to get something to work and represent a basic form. What makes the 3D Scene such an artistic challenge over photography everything is created nothing is given.

[Bibs]

That is fairly mind boggling.

[Malcolm Feth]

Great job! Seriously...

I always wanted to do some CGI. Of course - full focus on cars. Work on textures, modelling, rendering.

Modeling cars from scratch, using blueprints etc... cool stuff, but I would probably have to give up sleep to find time to do it (or work). Some of the cars renderings are so impressive and once you get your model you can re-use it in million different ways... this is seriously awesome.. ehh

[Iconic Ride]

Phil, I'm having some difficulties with the screen shots on my latest game console. Can CGI fix this?

Regards,

John

On a more serious note, how the heck did you get involved with this rather esoteric line of work in the first place? It's not like you can apply for CGI University and major in rendering. What sort of background enabled you to make the leap?

[OwenGT3]

Good read, I've been workiing with 3D for about 12 years now, but have never made a car, always wanted to, but never have the spare time to devote to the process. I suppose I'm what you defined as a 'generalists', as I do make a lot of images using 3D and then spend time reworking in 2D.

I used Studio Max, started on version 2, now on version 13. What's great about the software is I get a home license free from my works, so I don't have to buy it for home use, which saves me several thousands of pounds. The new A&D materials are excellent, also the day light system they introduced some years back. Rounded corners is another nice touch for making models look so real.

Some samples of my work.

This one has had some 2D Photoshop effects applied, motion blur and ripple

The closest I've got to a car, my own American Chopper design of my Darth Vader Bike

This is one I made quite a few years ago, but it's now had a make over with new materials (Below)

The only 2D work is adding a glow to the ceiling light.

As much as I love producing 3D work, I still favour making stuff for real, you can't beat it. The fact that you can touch and interact with it once it's made will always win over 3D work for me personally

I mean, to see a AT-ST going down your high street has to be seen! (This is real, not CGI!)

[mayevora]

Wow. Good read. Very interesting stuff indeed.

[Guest surferphil]

Great stuff owen!

John,

I left a poor school with nothing and there were no jobs all the factories had closed down (thankfully), so my mum sent me to Art college. I studied art and modelmaking at Uni before my short career as a model maker, building real models but could not live on the money and the chemicals made me ill.

I got a job as a junior and laughed at how cushy it was to work in an office, the pay was poor but I could live on it. The software, 3D Studio (before 3DSMax came) was awful to use but what it could do was much simpler, we sort of invented techniques and found solutions to problems as we went, much like art and real modelmaking but more technical. I think this is really how the industry was back in the mid 90's.

You can do a degree in Animation and major in CGI, but I haven't been impressed by most of the people who I have worked with from that background.

Anyway I still have a lot to explain and hope not to bore anyone away.

[Guest surferphil]

3D Laser Scanning

Having got to the stage I was with the model from scratch I stopped as I got hired to model the Volkswagen range for a company in London. They used a laser scanner to scan the interior and exterior of the cars and a small team of modellers (including me) rebuilt the cars using the scan data as a guide.

I was curious as to how accurate my techniques were using only photos as a guide compared to the 3D Scanning. I had become good friends with the two people who did the scanning and one night we were chatting and I asked if they could scan my car, they said you can and we'll tell you what to do.

So we scanned my car. Here are some pictures I took during with Leo setting up everything and supervising (picture below).

In the image below you can see the masking tape on the windscreen and white dots on the hood, these dots will be all over the bodywork when it is being scanned.

(In the image below.) The silver object next to the laptop is the Laser Scanner and the image on screen is the surface it picked up and generated the 3D geometry from. You can see it's picked up the masking tape and the fabric hood but the windscreen is missing and the side window is fragmented. This is because the laser can't get a focus on the glass as the light passes through or reflects (off still focussed). The scanner works best on fabrics as the laser is reflected back in many directions hitting the scanner and giving a reading.

The white reflective dots let the scanner software generate a map, if you stop scanning and start again the software will know where you left off and carry on building that same surface. If the dots were placed perfectly apart the software would not be able to do this, it would be confused as 1 triangle of dots would look the same as another to the scanner, it would be lost in the repeat pattern.

Unfortunately this scanner cant read a surface like we can, all it can to is plot points in space 1 at a time and it's the software on the laptop that generates the surface. Neither scanner or laptop know what the surface is or can use the data, it's a 1 way dataflow.

(In the image above) You can see the Classic Mini covered in Latex paint, the car has a metallic paint finish causing problems for the scanner so the latex gives a better surface to scan.

In the image below is the end result from the 3D Scanning. It's by no means a usable surface but it does help get the scale and shapes accurate.

You can see I had to cover the entire windscreen in tape to get a surface and only half the car is scanned as it gets mirrored when modelling. This can cause problems but not without solutions, asymmetric objects like the Dashboard and filler cap need to be scanned on both sides of the car. I also wanted the headlight pod scanned in the up position.

Below. The outside and inside close up; you can see the surface made of triangle polygons and the limitations of the scanner that require some artistry (luckily as that was my job).

Edited October 30, 2012 by surferphil

[Iconic Ride]

Phil, is this the same (or related to) the type of scanning (and associated software) that is used to generate "3D Printing" real-object production?

Is it possible to create a CGI set of data points from scratch (i.e, without scanning a real object) and plug it into a 3D Printer?

If so, could a sufficiently motivated graphics designer construct, say, a "virtual" Esprit using a sophisticated CAD program, and then use a highly advanced form of 3D Printer (in the future) to "build" the car?

What is the relationship, if any, between CAD and CGI?

[Guest surferphil]

There is a direct connection between CAD Data and the type of data I use to generate still and moving images.

I frequently use CAD Data to create still images.

I will come to that in a bit

I can create a model for 3D Printing but there are (simple) certain criteria you need to meet in the way objects are arranged in the file. It is best to rebuild this type of scan data into clean closed objects as scan data is so rough it will confuse the printer.

There are other ways of scanning that produce a cleaner model but they are laborious and don't lend themselves to scanning such a large physical object such as a car.

There are also many types of 3D printer the best uses a resin bath where a laser cures the resin in tiny points that bond together or it turns a pint to dust so you can get undercuts and hollow sections, ideal for mechanical parts like an engine block where there are hollow water pipes and complex surfaces, this is 'rapid prototyping' and a finished piece can be delivered in a couple of hours from start to finish. The real world analogy would be sand casting (if you understand that).

[Iconic Ride]

"The real world analogy would be sand casting (if you understand that)."

Of course I understand that, Phil. I've built tons of sand castles, but mostly when I was a younger man.

Just kidding. Is sand casting anything like "lost wax" casting?

[Guest surferphil]

Yes but the analogy is only intended to illustrate the result not the process i.e. The complex model you can cast in 1 single piece of material. On reflection, it was a poor analogy.

CAD Data

Most manufacturers I have worked with (Ford, Volvo, Mclaren, Bentley, JLR, etc.) provide CAD Data, this comes from the 3D Modelling departments or engineering companies who use 3D models in visualisation, rapid prototyping, computer controlled milling machines and virtual reality to aid design development. The data is used by machines and robots to make tools (moulds) and machined parts in vehicle (and other product) manufacturing.

The CAD models differ slightly to the CGI models in the way the data is formed. CAD uses patches, grids made of mathematical splines all linked together to make up the surface of an object.

To get the CAD to work for CGI, the spline patches are translated into triangulated geometry (Polygons) either directly or indirectly through an intermediary translation package (such as RTT or Polytrans) where it can be imported into a (CGI) 3D software package. The result is much like the scan data but with much cleaner renders, the data from the smoothness and curvature of each spline patch in the CAD, is retained during translation in a hidden dataset called vertex and face 'normals'. This data is usually locked so the computer does not smooth faces automatically like it would with my modelled data shown earlier. This allows for some pretty shocking geometry to look really good when rendered.

Without the normal data the geometry would look as bad as the scan data.

Where possible, CGI production utilises existing CAD models (and architectural drawings) simply to save time. These parts will be signed off for final production on a vehicle, so they will be exactly the right size and shape in the artwork as they are on the finished vehicle. A full CAD model of a car consists of such details as; electrical contacts, printed circuits, door lock components, pistons, bearings and screw threads. All modelled exactly how they will be manufactured.

The limitations for using CAD data are easy to overcome, but there are a few problems inherent with using these 3D Models:

Soft objects like CV boots/gaiters and wires are tricky as they are bent during a real cars assembly, but come out of the mould strait, so the CG models need to be manipulated by the 3D Artist or modeller preparing the data to be used for CGI. Most components need to be stripped out as they take up resources slowing down the PC with redundant data.

You get every single component that is required for a real car included in the 3D models of the adhesive used to bond panels and glass.

Below. The interior parts made of 4 sided faces were modelled and the interior parts made of triangular faces have been translated from CAD Engineering data. The model gets very dense and difficult to distinguish between translated CAD objects and you can see how clean the modelled data is compared to the translated CAD.

Range Rover Evoque interior.

A full car model in CAD will be unusable on a powerful PC when translated to a CGI model because the data translated holds so much more information than CAD. Other objects like leather seats are often missing, the foam that makes up the seat is the wrong shape, it's shaped how it comes out of the factory mould, rather than the final shape of the seat in a showroom car which has been squashed into the stitched up upholstery cover.

Because accurate CAD Data is used for manufacturing the car, objects will arrive complete with the back of a panel you don't normally see in a real car (unless you take it apart). The back surfaces can have a huge amount of detail much more than the surface you see. On a large job this can be stripped out to make a much more simple model that will render quicker and enable a computer to be used a lot quicker by an Artist working on the model.

Below are examples of a rear passenger door that has had all the objects you don't see removed, but still all the back surfaces remain, slowing the job down slightly. It takes a lot of work to remove the un needed faces, so as a manager it's important to decide weather removing superfluous data will benefit production over all. In this example they remained.

Below shows the data you don't need for CGI but it's there for engineering.

XF Sports Brake interior.

Edited November 1, 2012 by surferphil

[Guest surferphil]

Texture Mapping and materials.

 

CGI can often look flat and plastic without the fine natural detail abundant in nature and the image in this example would not work without the detail in the environment being reflected in the car body.

 

To create the reflections in materials like car body paint and glass I used some edited photographs of sky and landscape to help visually seat the car in the environment.

 

Typically texture mapping is used to add fine detail and create the illusion of textured surfaces. Decals, natural materials and surface textures can be applied to parts of a model. Using a jpg image (or other image formats) a piece of geometry can be mapped so the image will appear in its desired position. This is exactly how computer games look so good without grinding to a halt trying to calculate almost infiltrate detail.

 

Below is an example of some of the images applied to the model in this scene.

 

 

 

 

The image of the sky is created in Photoshop from many photos:

 

 

 

 

The environment image is applied to a dome and reflection properties are set in the car paint and glass materials to achieve the appearance of a glossy finish. Material properties can be set in the 3D package (it can take days to perfect the look of a single material).

 

Below is a screen shot of how the model appears with the materials, it's not a simple process and the results can only really be seen when the image is test rendered, not in the work space where the changes are made, making it difficult and time consuming not being able to see what you will end up with.

 

 

The glass and painted body panels have been rendered separately to save time and gain more creative control over appearance and transparency of those materials. Because the bodywork reflects back light and images at different angles the environment distorts just like it would in real life over the curves and angles of the car body. It is essential to do this in the 3D package but tweaking things like colour, contrast and transparency is a lot quicker in Photoshop where the results of any changes are seen immediately.

 

Below are the different images rendered so I can control each one on a separate layer in Photoshop.

 

 

 

Below are the layers composited in Photoshop Elements 11.

 

Edited January 4, 2013 by surferphil

[Bibs]

I found this interesting Phil, CGI masquerading as non-CGI! 

 

http://www.fastcodesign.com/1671941/why-this-oscar-nominated-disney-short-looks-like-nothing-made-before#1