Modellering och shading för reklam och film

BACHELOR THESIS

Asset Creation for Commercials and Film

Stefan Berglund

Bachelor of Science

Computer Graphics Arts

Luleå University of Technology

Luleå Tekniska Universitet C-uppsats

Datorgrafik

Asset Creation for Commercials and Film

Stefan Berglund 891103-1519

steber@hotmail.co.uk 2011-06-02

INDEX

5.4 Texturing and Shading 5.5 Conclusion

1. ABSTRACT

The thesis describes the experience of my internship at Fido Film, Stockholm, and a selection of the tasks I worked on. This is done to study the performance of the work done and detect problem areas, especially when compared with industry requirements. The internship took place over nine weeks in April-May 2011. I worked mainly with asset creation on a half-dozen commercial and film projects. Particular focus is placed on describing the main three tasks I worked on: a fish, a cartoony character, and a cleaver. Together, they comprise over one third of the work hours I spent at Fido. In general, the transition from controlled educational environment to an actual workplace was trouble-free. Better understanding of the properties of real-world materials and the shading models built to represent them would have been beneficial, and it is suggested that the graphics’ programme place a greater focus on reviewing progress to improve the students’ awareness of their own and others’ work.

2. INTRODUCTION

2.1 The Industry

Commercials and films have a continuous need for assets in their productions. Companies

specializing in providing high-quality stock models have gained ground in the industry, though most productions still require custom-tailored assets for that particular project. Thus both a high level of detail and realism, and adaptability to the project environment, is required.

In smaller companies the process of producing an asset is the responsibility of a single person skilled in several areas, while in larger ones, modelling, texture painting and shading become more and more separate tasks to capitalize on the individual’s expertise in a particular field. To investigate the nature of the profession of asset creation, then, the more precise requirements of the large company informs of the demands of the particular fields, while a smaller company allows one to actually practice the full process of asset creation by oneself.

2.2 The Requirements

To quickly review the requirements of the large company, then, one can find them in written form on job listings. On The Moving Picture Company’s website (2011) the general requirements for all jobs in asset creation demand:

- A deep and demonstrated (in previous work) understanding of real-world materials, a traditional arts background being advantageous, and knowledge of the most relevant software being a necessity.

Also required is to actively “improve your knowledge of MPC’s pipeline”, and for senior positions, to have “proven technical and creative ability in a broadcast or feature-film production environment”. Participation at dailies is expected, and focus is placed on a structured workflow which allows adaptation to feedback and design changes. One must effectively communicate and problem solve, and be knowledgeable about the other disciplines of asset creation.

For modelling, it is added that work will range, stylistically, from the realistic to cartoony; that proficiency in creating deformation-friendly meshes is needed, as well as skill in interpreting models from 2D reference and 3D scans. Finally, a “good understanding of the entire visual effects process” and a desire to improve is called for.

Texture artists are required at MPC (2011)* to be able to paint “photo real textures and relevant maps”, either from scratch or from reference, to create new objects or to match existing live-action material. On Framestore’s career website (2011) Shader Writers needed to have strong artistic lookdev ability and knowledge of technical issues, such as HDR lighting, colour spaces, BRDF and other rendering principles required for physically accurate shaders.

2.3 The Internship

While some of these requirements are outside the reach of any academic education, it is the obvious goal of the education to prepare students as well as possible for production demands. An internship provides the otherwise inaccessible demands of the hands-on experience of the actual industry and its workflows, as well as raising the demands on the student’s results. Thus the performance of the student during his internship is a benchmark of his own readiness for the professional environment -and his education’s success in meeting the dem-ands of the industry. The internship is, then, not only important for the student himself, but should also be the occasional check for the Education itself.

I applied for an Internship in asset creation at Fido Film in Stockholm. It is a mid-sized company focusing on creating creatures - animals - and other content for commercials and films, and keeps a high quality and standard of realism.

The purpose of this thesis is to describe the work I did at Fido Film, and to relate my experience there to the requirements of the VFX industry and to the education I received at LTU. By describing, in some depth, the process behind some of the more demanding tasks I performed, I hope to have a solid basis to review my performance - my success, or lack thereof - in the various fields of asset creation, and the reasons for it.

3. METHOD

3.1 Working at Fido 3.1.1 Employees

Fido has about 25-30 employees, and a handful of freelancers. The company is divided into four departments of roughly equal size; the administration, the 3D, the compositing, and finally, the technical, R&D and FX. There is also a - for the time unused - workshop.

Fido, being a part of the general movement in the VFX industry to informed and self-aware artists that dare make demands of their working environment, is among the first VFX companies to introduce pension plans for its employees. At the end of May, Fido Film was bought by Forestlight Studio, with which it had collaborated previously.

3.1.2 Organization

Fido uses Linux as its main operative system, though for particular tasks, some used Mac or Windows for easier and more stable workflows. All work is done and stored on the network server called “flock”, which is backed up continuously. Fido uses - and develops - Ftrack, an in-house management system accessible via the browser, to be used more and more in step with the file structure in flock and for administrative duties. Due to the evolution of ftrack and the changing needs of Fido’s projects, the standard folder structure created for each project at start-up had been

changed a short time before my internship began. In aiming for a “universal” folder structure, the most recent iteration was slightly confusing, and not everyone was quite sure where what went, or was supposed to go.

3.1.3 Openness

When I started at Fido, I was assigned a Fido Buddy to help me with any questions I had. But it was soon apparent that any particular Buddy-system wasn’t needed after the first day or so; while Fido is one of the larger VFX companies in Sweden, it is still a socially open environment. Whenever I had a question, I could ask whoever was closest, or my supervisors, or go to someone knowledgeable about my problem area.

Fido has a Monday breakfast-session, during which the progress of the past week is reviewed, and the plans for the next one are summed up. Further, dailies for each project is kept usually every day, so it is easy enough to keep track of the major comings-and-goings of the company as a whole, as well as how things are progressing for the different projects one is involved in. Even when the information isn’t - strictly speaking - necessary for performing your tasks, it helps keeping an open and informed air in the company.

3.1.4 Software used

Maya 2011 was the main tool used for most of my tasks. Base modelling and shading was done exclusively in Maya, though Renderman shading was carried out in Slim.

Photoshop CS2 was used for most texture-work. zBrush was used for a few projects to 1) improve low-res models, 2) create high-res models for displacement maps and 3) some texture work. Mudbox was also used, mainly for texture painting, as I found it more flexible. Both Photoshop, but more so ZBrush, were noticeably Windows-programs used outside their native environment. Simple tasks like navigating through the (rather large) folder structure were a bit frustrating, as one couldn’t really bookmark anything or copy-and-paste addresses to the address field or use other shortcuts. Also, along with the recent changes in the folder structure, no ZBrush had no plug-ins.

Nuke was used by me mainly as a tool for lookdev and preparing material for dailies, but also for some actual compositing for projects.

Jobs could be submitted to the render farm from both Maya - Mentalray and Renderman jobs - and Nuke.

3.1.5 Pipeline for me

Mostly, I was assigned to new jobs whenever I was done with my previous task, in a sort of drop-in schedule. Occasionally a Supervisor had something they needed done at once, or I was informed a few days ahead of an upcoming project I would participate in. Often I worked on several projects simultaneously; while waiting for feedback on one task, I would start up on another.

First, of course, I was introduced to the task: what it would be used for, and in what further context. Most of my tasks were to recreate an asset from reference: sometimes with the actual object at hand, or from concept/photographed reference; occasionally also filmed reference. As much reference material as possible would be made available, and a short discussion of the method of execution would be held and practical concerns addressed - polygon count, deformation-friendly topology, etc. How it would be used in the pipeline - would it be rigged, for instance? I would, as a rule, also be pointed to a more direct supervisor to help me in my work - for modelling- and shading-questions, rather than ones to do with the task as such and its use in the project. With the folder structure generated, I could begin working and log time in ftrack against the “bid”, the planned time for a task.

Almost all tasks I received were either “Modelling” or “Lookdev” ones; texture-painting and shading is treated separately - often I would eventually do both, but not directly together.

Nearly every day, each project has a daily-meeting. Generally, for modelling and lookdev,

turnarounds were rendered out and reviewed, and for tasks more directly related to a particular shot in its entire sequence, well, the whole sequence would be rendered out. The discussion kept at dailies range from commenting on the technical properties of the material reviewed - “higher

reflectivity on the eyes” - to more general comments on the direction of the project and the pipeline; simply planning the work to be done.

Depending on the size and complexity of the task, I could submit several dailies before a final approval. Most tasks didn’t take too long - two, three days - while others took a week or two in effective work-time. When receiving a final approval, I would make sure to clean up my work files to pave the road for the next step in the pipeline.

3.2 My work at Fido Film

Many of my tasks were related to the German feature-film project Yoko, directed by Franziska Buch, with Hanna Bengtsson as Line Producer on Fido. Fido worked together with Forestlight on this project. I worked on half a dozen assets, where the simplest was an afternoon of setting up a Renderman shader from already-painted textures for a ball to be used in the background, and the most time-consuming was a frost-covered fish to be viewed close up, along with a partially-eaten variation of the same fish.

I also worked on a handful of other projects, creating a crackers package, modelling a character to be animated for a infomercial - and set up a Nuke comp for the same project; I organized and optimized about a dozen sets of models purchased in bulk for a commercial, hand-tracked a shot for the music video of Kate Bush’s Deeper Understanding, and did some preliminary 3D-work on a tree to - eventually - be finished in 2D.

I will describe in some detail the process of the three largest tasks I worked on.

3.2.1 Fish and eaten variation

The task was to recreate a fish prop that had been used at the shooting. In the film, it is held up relatively close to the camera, and is covered with frost from the refrigerator it had been in. It is then partially eaten. A bit later in the movie, there is an actual, partially-eaten fish shown.

Thus the asset I was to create had to match most closely a plastic prop covered with a chalky substance, available to me at Fido, but also an actual frozen-and-partially-eaten fish.

Figure 2. Photographed reference for asset creation

My first task was to model the whole fish; the eaten variation and texture/shading (lookdev) were all later tasks, potentially to be performed by other people. I photographed reference myself, trying to get orthographic views.

The modelling was straightforward, and done entirely in Maya. Because the fish’s body is curved the orthographic views didn’t quite suffice - the top and bottom view being rather botched - the main comments was about the length of the head, and of the roundness of its “moustache”. Supporting edge loops were added in as the model would be smoothed. It was decided that it wouldn’t be actually deformed, being - after all - frozen and the transition between “whole” fish and “partially eaten” being mostly hidden from the camera. I kept the topology to quads; the final count was about 10,000 triangles. When otherwise done with the model, I created a simple joint-chain, skinned the model to it, and used it to curve and scale segments of the fish’s body to match the shape of the prop. The uv-map was split into two shells, dividing the fish along its length.

Fig 3: the finished, smoothed model.

The partially-eaten variation was to be a exaggerated, comic version, of the classical fishbone-with-just-head-left variation. The already-made model was chopped up into two pieces; the head and tail portions were kept. The middle was replaced with a spine with little bits of meat sticking to it. For the spine, a single vertebrae was finished an uv-mapped before being duplicated to form the whole spine. The bite surfaces were sculpted in ZBrush to suggest the forms seen in the film’s fish. (There existed no reference of the eaten fish beyond that of the actual footage itself.) The partially-eaten variation was modelled - and kept - straight, as it was to be deformed in animation.

The uv-shells of the partially-eaten fish was fitted into the uv tile of the whole version, creating a single atlas map for both assets.

I worked in 4096x4096 resolution, and kept most all work in the same .PSD file - creating, at times, files well over 500 mb in size - and saved out .TIF files to be used.

The first step was to project the reference photos I had taken onto the model (the whole version) from a couple of different views, then transforming them together in Photoshop. For the projection, I simply created a new shader with the file texture as a projection, manipulated the projection

controller, and, in the Hypershade, Converted to File Texture. I focused on one side of the fish for a long time in the process, leaving the other until the shading of the first had been ironed out.

I then, following a rationale of the quick and dirty fixes, I made a few maps for a blinn shader, shown in figure 5. Eccentricity was a dimmed-down version of the diffuse map, specularity an inverted and colourized version, and the bump was a slightly modified diffuse map sent through a PS2 height pass filter. A rough mask for the frost was created.

Figure 5. Clockwise from upper left: 1) diffuse 2) eccentricity 3) specular colour 4) bump

Figure 6. A Mentalray render

At this point, I presented my progress at a daily and got several comments. The most important of which was that the fish was to be shaded in Renderman, using Slim (which is, simply put, the Hypershade of Renderman). A more methodical system for the shader was suggested; to create a base-specularity layer to emphasise the individual scales, and a thin coating of fish-sliminess above, with the frost’s specularity being a third layer. And, a better bump was required.

The first thing I did was to return to modelling. In fact, I pushed the fish into ZBrush and sculpted every scale, using the colour-texture I had created as reference to place them.

Figure 7. ZBrushed fish.

Next, I made a few new maps and edited others. I made a thorough paintover of all frost, mostly to use the Difference between the layer-with-frost and the layer-without-it to create an accurate mask for the frost itself. I set up the displacement and three spec-layers in Slim - which took some getting use to.

Figure 8. Clockwise from upper left: 1) overpainted frost in colour, 2), 3) and 4) the specular colour of the three layers

Figure 9. A first Renderman render.

The Renderman shader wound up with three layers of specularity. The scales’ specularity, which used 2) from Fig. 8 as colour, about 0.5 in scale and a low roughness of about 0.15; the overall specularity,

using 3) from Fig. 8 as colour, a scale of 0.2 and a roughness defined by 2) from Fig 5. Finally the frost layer, with a scale of 0.6 and a roughness of about 0.1.

A few extra shading components were added in, such as translucency, and a couple of diffuses with a scale set to 0 to create AOV outputs - that is, extra colour passes, such as 3) in Figure 10, to control the image in compositing.

When most all other shading was done and textures painted, I projected the texture maps from the whole fish to the partially eaten version (Rendering>Lighting/Shading>Transfer Maps> 4k, medium quality). The fleshy bits were textured with projections in ZBrush.

Figure 10. Clockwise from upper left: 1) frostier diffuse 2) roughness of overall spec 3) AOV map 4) extra frost bump

Figure 11.

Some further tweaks remained. More frost - especially along the fish’s top - was added, and a Rim shading component was added, broken up with a modified version of 2) in Figure 8.

Figure 12. Large version shown in Results.

3.2.2 Character for commercial

The second major task I worked on was a character; a crown with a face and limbs. I was given finished concepts in orthographic views to work from, but there was still a good deal of decision-making in realizing the 2D concepts into 3D forms, and balancing realism with stylism to create an appealing character.

Figure 13. Character concept.

I made a fairly detailed base mesh in Maya, to make sure to adhere to the concept as closely as possible. The crown - and, to a lesser degree, the face as well - proved quite impossible to follow precisely while still maintaining a round shape of his body. At first I decided to go with the

orthographic views, pushing and pulling the originally perfectly cylindrical body to match; I even got started in ZBrush on a rather loosely shaped sculpt.

Figure 14. First, free-hand, version of the character.

However, after a discussion with my supervisors, I retraced my steps to make a more symmetrical body. The crown was remade by first modelling on straight strip, that was deformed into a circle. His face was cut out and a perfectly cylindrical body was added in, the face-covering part being masked out. This made it easier to control the transition from the face to the rest of the body.

Most of my feedback had to do with the shape of his eyelids; it was tricky, trying to translate the extreme cartoonism of the concepts into 3D. He was modelled slightly smiling, partly to simply “sell” the character to the client.

Figure 15. Sculpting near done.

The shoes were first sculpted in ZBrush to get a sense for their forms, then carefully remade in Maya. When the sculpting was otherwise approved, the face- and body-parts were projected onto one mesh. This was eventually retopologized (and so the detail was re-projected) as it was decided to not use displacement maps, but rather increase the resolution of the model in the face and simply smooth it.

I decided to create the new topology in a somewhat lazy way, which turned out to be quite time consuming. The base model was about 10,000 triangles, and to comfortably animate it, it shouldn’t exceed 30,000. I chose to simply subdivide all the faces in the face region, rather than inserting single edge loops one by one. However, this caused smoothing problems, and I eventually had to resolve all the edge loops by hand anyway.

Figure 16. Clockwise from top left: 1) Original, hardly 10,000 triangle base mesh. 2) Partially subdivided. 3) Smoothing problems - “goose bumps”. 4) Final mesh at about 20,000 triangles.

3.2.3 Cleaver

The final major task I worked on was creating a cleaver for the Yoko project. The task hadn’t been planned for while shooting, but come up during editing. From reference in low-resolution footage from the scene, I had to create a model that would hold up relatively close to the camera.

Figure 17. Technique to model the cleaver

To have something to base my scale on, I used a shot of the cleaver lying on a table, and positioned a cube to represent the table and match its perspective. By projecting the image of the shot onto the cube, I had a foundation to work upon.

As metal is highly reflective, it was decided to shade the cleaver in Mentalray. The mia_material was used for both the wood handle and the metal blade. They objects were tiled into the same uv space and shared textures, but not shaders. Both handle and blade was uv-mapped split along the bottom edge and unfolded in one shell.

The projected colour was converted to a file texture to aid in colouring. For the wood, I simply used a rather grainy texture, and rotated it to reflect the cuts in the surface. For the metal, I sampled several different leaky-metal textures - as the blade in the shot has some kind of dark streaks I took to be blood or just discolouration - and merged it together.

As with the fish frost, I soon enough painted over the leaks to produce an accurate mask of the leaks themselves. Scratches were made and kept assiduously on a separate layer. The wood was given low reflectivity and glossiness, but a patch of higher values at the very grip to imply a surface worn smooth from usage. The glossiness and reflectivity maps are almost the same, the main difference being the value of the blade’s dark streaks relative to the metal itself.

Figure 19. Clockwise from top left: 1) diffuse 2) glossiness 3) reflectivity 4) bump I was helped in setting up the scene, and a useful tip was to angle the blade slightly - the surface properties of metal are hugely dependent on reflectivity and angle of incidence. Proper linear workflow was established, and the Specular Balance property of the mia_materials was set to 0. Further, an env_blur node was connected to optimize and improve the near-exclusively environment reflections.

Figure 20. Second daily.

The comments here were mainly that the dark streaks were too pronounced, that the wood should be darker, and the edge of the blade cleaner. I tested a few variations, and finally switched my reflectivity and glossiness maps - making the dark streaks less reflective and more diffuse than the metal around it. I also set up a custom colour pass, to provide masking control in compositing.

Figure 21. From left to right, up to down: 1) diffuse 2) glossiness 3) bump 4) custom colour 5) third daily.

The comments were mainly that the blade could do with more reflectivity - or lower diffuse, at least. In fact, pure metal does technically not have any diffuse component according to Wikipedia (2011) though impurities and surface layering counters that. While the diffuse weight had been set to about 0.2 all along, it was now reduced to about 0.1, and along the polished edge, a very low value of 0.02-3. Further, the BRDF 0 Degree Reflection value was increased to about 0.7.

4. RESULTS

Fish and partially eaten variation: time for Modelling: 3 days. Lookdev: 7-8 days. Number of full, 72-frame turnarounds rendered: 10+.

Figure 24.

Figure 25.

5. DISCUSSION

Overall, my internship passed without much trouble. The comments from my supervisor, Anders Singstedt, was that I seemed to have coped well with the tasks assigned to me, and finished them within the given time limit. The main exception was the shading of the fishes, which required me to acquaintance myself with both Renderman and Slim’s shading system before any meaningful progress could be made.

I will end with going through the Requirements mentioned in the Introduction and briefly discuss how they have been relevant for my internship.

5.1 Personal skills

With the exception of Fido’s using of Linux and ftrack, all programs I used - Maya, ZBrush,

Photoshop, Nuke - I was already well acquainted with. I briefly used Mudbox for some tasks, mainly texture painting; a deeper understanding of that software could, no doubt, have helped for a smoother workflow.

While my education at LTU has provided me with some traditional arts background, helping me in design choices, an understanding of real-world materials was more obviously required in most of my tasks, and I felt sorely unprepared. More about that below.

5.2 People skills

The pipeline consists, to an extent, of material being passed on from one artist to another; it is also a constant discussion between the artists of how to best and most effectively deal with the material in question. It is a requirement to have a well-working pipeline to quickly deliver high-quality material; a pleasant side effect of this is that everyone involved are generally helpful and open-minded. Feedback seems less a reflection of personal performance than a neutral observation of what needs to be done.

At LTU, the education has successfully promoted working in teams, but a main ingredient is still missing; that of the constant reviewing and feedback on progress, and interdependability with other artists. An asset isn’t necessarily done just because it looks good; it must also look right, and it must also be able to be used by someone else in his pipeline. Discerning what can be improved in others’ artwork is a helpful way of improving your own work as well. Dailies are a powerful tool used throughout the industry; it could be well used - in some form - more prominently at the graphic’s programme.

5.3 Modelling

The modelling stage - especially for characters, that go on into a complex pipeline - is undeniably crucial to get right, else all work after it will be affected. The character I created was interesting in its precise mix of realism and stylism, which it took some discussion to get quite across. The most frustrating part was the topology and trying to make it deformation-friendly.

5.4 Texturing and Shading

From the fish and cleaver projects, it has been made clear that any sophisticated shading is based on thorough and structured texture paintings. When painting otherwise, I prefer to keep the number of layers as low as possible, working as if it were an actual canvas; but the necessity to be able to access and edit particular parts of the texture map was well demonstrated by the frost of the fish and the streaks on the cleaver.

Further, it has been driven home that shading obviously requires a lighting setup, and most preferably that in which it will be used. That last part wasn’t available too often during my time at Fido, for the obvious reasons of what comes first in a pipeline. The importance of a linear workflow to create convincing images has been a recurring theme as well. Strangely, though I’ve read and been instructed several times in the ways of applying a linear workflow, I’m still hesitant about it - it simply looks (and works) in a different way to what I am used to.

Finally, on the texture and shading, I have experienced the necessity of actually understanding the models and attributes available in the shaders, how they relate to real-world properties, and how different real-world materials are built up. The low diffuse weight of metals is a good example - blind tweaking of attributes might happen upon favourably-looking results, but for consistent and

convincing materials, there is little doubt that a foundation in fact is a must.

5.5 Conclusion

While the internship offered a few challenges, novel requirements, and a different workflow from that used at the computer graphic’s programme at LTU, overall, the transition was not too rough, and seemingly successful. For asset creation, a deeper understanding in physical properties and how they are represented in shading models would be beneficial, as well as some further instruction in proper rendering workflows. Finally, reviewing each other’s progress is far more instructive - for both parties - than commenting on a finished work; this should be promoted at the programme .

6. LITERATURE CITED

F

Framestore 2011, Framestore Careers, viewed 29 May 2011,

<http://careers.framestore.com/job.html?job_id=128&location=London>. T

The Moving Picture Company 2011, RECRUITMENT - Vacancies - London - Film, viewed 29 May 2011, <http://www.moving-picture.com/index.php/recruitment.html>.

W

Wikipedia 2011, Diffuse reflection, viewed 29 May 2011, <http://en.wikipedia.org/wiki/Diffuse_reflection>