• Robert Doman

Audi Quattro Breakdown and Retrospective


In the future I would love to be a 3D vehicle artist and when starting this assignment, I knew I needed practice. In my previous models I didn't aim for too much detail. I knew I would need to improve my capabilities and began research on a new vehicle. It was important to chose something I could find a lot of reference on in order to create it accurately. My inspiration began last year when I bought Dirt 4. This gave me a bigger interest in rally cars, leading me to the iconic 1980 Audi Ur Quattro.

My first real car model was the Blade Runner spinner, followed by a 1969 NASCAR. These were both mid-poly, so I wanted to create a vehicle without using decals or floating geometry to create textures. Looking at these Codemasters job requirements, I wanted to make sure I could prove I can model and texture 3D vehicles. For my final outcome I planned to model and texture an Audi Quattro exterior using Maya, Substance Painter and render it in Marmoset Toolbag. I wanted to learn Marmoset to boost my rendering skills.

Seting Up

I found the best source of visuals for this vehicle online. I wish I had tried to find someone in Cornwall with this car. This would have allowed me to take first hand reference and clearly understand the details of the car.

My main frame of reference for modeling has been Karol Miklas. Their Sketchfab models helped me to visualize how to approach high detail vehicles and unique texturing. A 80lv article featuring Karol also gave me a deep insight into parts of the production I had not yet discovered, such as planning and optimization.

1981 Audi Quttro from ipocars.com

Research into the car was continuous. Whenever I felt stuck, doing more research and finding more images was always helpful. The main website I used to model the outside of the car by was from ipocars.com. This is because the pictures are of the car at useful angles and are high resolution. Although the model shown is from 1981, my research told me it was the same as the 1980 version.

1981 Audi Ur Quattro outline from the-blueprints.com

The-Blueprints.com provided me with the blueprint wire frame of the exact Quattro from ipocars. This gave me the solid foundation I decided to build from. I quickly created a blockout as I have done with my previous work. After this, I created a Polycount forum to track my progress and get feedback.

I also found it useful to look at models of the car other people had made. These being cgtrader and 3Docean. Other sources I used included Classic Audi and Italeri for the colours and rally aspects. VWVortex is a restoration blog full of Quattros from many angles and taken apart. These helped me to build up the model and it’s detail.

The Process

I began with blueprints and block shapes. This is extremely useful for finding the form and feeling of the car very quickly. However, I found it fell apart when I needed more detail. Through advice I received through forums, lecturers and articles, I soon realized I should be using Nurbs to create the shell.

Nurbs of Audi shell used for creating the body.

I attempted to use Nurbs during the previous summer but I found complex shapes to be extremely difficult. After rebuilding curves to have the same spans, I would use the boundary toll or loft to fill the space between the curves. This technique still holds it's place but I learnt it was not for everything. I was shown the polygon tool and my modeling changed for the better. This tool let me draw polygons while snapping to curves to create my own custom shapes without tediously rebuilding and lofting curves. This one single technique was the best and most useful skill I have learnt while working on this project and without it I think I would have taken twice as long and lacked the ability to create the overall continuity I managed to finally achieve.

I also learnt that despite a car is in pieces, the shell should be modeled as one. This keeps the fluidity of lines and light bounces consistently. I also devoted more time to edge flow and as a result, I have a greater understanding of what makes good topology. Although my model has imperfect areas, I know to plan ahead and keep quads continuous around edges. I aim to practice this as much as possible from now on.

One of the first things I modeled was the wheels. My research taught me that there are three main types of tire pattern: symmetric, asymmetric and directional. I decided to create a symmetrical pattern so that I would only have to duplicate it and rotate it to create the wheel on the opposite side. Using a tutorial and a few attempts, I learnt that the design has to be tiling and have a flat end in order to be properly bent into a cylinder using the bend tool. This was a very useful technique as I could create the tread on a flat, easily editable plane. This would then be duplicated and bent meaning I could go back to the original easily. These wheels were extremely high detail, so I ended up baking them down onto a low poly. I used stacked shells to optimize this. I was very happy with the outcome.

While modelling, I began to feel the limitations of blueprint images. At a low quality, there is no hard line, so I had to make a lot of guesses. I used reference images a lot to make these assumptions but I know I still made mistakes, such as the bonnet being slightly too raised.

I focused on modelling the outer body until I was satisfied. Even after that, I kept finding little things I had forgotten, like the exhausts. It was important that I worked non-destructively so that I could address these issues easily. Towards the end I found it difficult, for this reason I always kept older versions of my work. One of my major observations I had is it doesn’t have to be perfect from the start. It is inevitable that change will occur. In the beginning it is vital to find the correct scale, form and flow of the shell in order to build from. As people suggested to me in my polycount forum, using a dark, high specular material will help you spot any issues while modelling. Keeping the tri-count low also means that you can more easily make changes while modelling.

Rear of car, rendered in Substance.

I textured the shell in Substance Painter. The rest of the materials are mid-poly. I enjoyed using Photoshop to create the logos to stamp onto the car. I sliced my high-poly to apply the painted areas in Substance. Using an ID map from my high poly model I could easily make changes to my UV map and reapply the ID map. This sped up my work as whenever I re-imported my low-poly mesh, I didn't need to re-paint anything.

I rendered the car a few times while I was working on it (mainly in Unreal), to test the model and experiment with the environment. I wanted to push the final piece by creating a realistic environment to place the car in to make it look like a true rally car. To do this I would use Megascans, a library of photo-realistic assets free from the Unreal marketplace. However, I could only download Megascans at home due to the version of Unreal at Uni. Unfortunately, I found my computer could not handle them as effectively as I had hoped. I came to the conclusion that I would have to render it on Uni computers. I still managed to render the car in Unreal but in a Studio scene.

Real-time render taken in Unreal Engine 4

My main takeaway is there are many different approaches for everything (modeling, texturing and rendering) and mixing them is the smartest way to work. Ultimately, whatever gets the job done quickly and efficiently is probably the best way to go. However, it is always important to use care and effort and not cut corners, which will affect your intended outcome. Recreating something perfectly is an ongoing process and strong reference is crucial. Furthermore, feedback proved to be useful for improvement. Showing my work to someone who has already gone through the same sort of challenges helped me to adapt and avoid making unnecessary errors.

Final Product

Overall I feel I did what I set out to do. I significantly increased my knowledge of creating correct topology and modelling high poly cars for texturing.

I also learnt how Nurbs are used in the industry and how to work with them effectively. My UV unwrap seemed to be correct. I kept similar areas together, but I think I could have reduced the space in between them. I think the livery looks almost indistinguishable from my references.

I am glad I managed to render it in Substance Painter and Unreal but I wish I had done what was necessary to make it look as though it was on a track. I learnt about real-time rendering in Unreal more and feel the final shots look very real from a distance, but start to fall apart under scrutiny.

Left: My render. Right: Historic photograph of an Audi Quattro rally car.

I know I overlooked a few aspects: the fuel cap, lights, bonnet vents, wheel bolts, antenna, windscreen wipers, 'Audi Quattro' windscreen logo, engine and believable underside, interior detail, brake pads, detailed rally lights and fender flares. More elements will probably appear as I continue working. I have a few baking and ID map errors. These are due to my high-poly which I would like to recreate.

My Plans

I want to continue with this car and address the problems to improve professionalism. I would also like to use vertex painting in-engine to add dirt to the car. I want to create visual effects to make it look like the car is kicking up dust as it drives. I will make the environment around it using Megascans. I will do this over Christmas since I can do it on my computer as long as I don't overwork it. Although I originally wanted to render the car in Marmoset, I didn't focus on it and so didn't get time. I would love to learn how to use Mamoset, and will to next Semester.

I also want to start a new car from a different era and focus on the engine and power-train. I want to do this to practice more intricate hard surface modelling. I will probably re-create an old engine since they are 'simpler' and don't have plastic covers hiding the details.

In conclusion, I'm extremely happy with what I created and want to continue pushing myself to achieve more.

To see the model on Artstation, click here. To view the model on Sketchfab, click here. To see the original Polycount thread, click here.