Stylised rendering in a competitive multiplayer game is a harder problem than it looks, and it is a problem of two conflicting requirements. The art has to read as illustration — comic-book, brushstroke, graphic — and the gameplay has to stay legible at speed with six characters’ abilities firing at once.
Marvel Rivals’ art team walked 80 Level’s David Jagneaux through how they resolve it — seven team members including Art Director Dino Ma, Lead Character Designer Yuzhen Zhang and Lead VFX Artist Peng Weifeng — with Cyclops as the case study.
The pipeline, in order
Research across comics, films and animation → concept art → modelling and texturing → rigging → animation → VFX implementation.
Worth noting that research across three media is the first step rather than reference-gathering during concept. For a licensed character with sixty years of inconsistent visual history, deciding which Cyclops you are making is the first real decision, and it is an editorial one before it is an artistic one.
The hybrid NPR/PBR shading system
The team describe combining Non-Photorealistic Rendering (NPR) with “the highlights of Physically Based Rendering (PBR)” to get stylised, 2D-like brushstroke effects inside 3D environments.
That hybrid is doing something specific and it is worth understanding why you would want it.
PBR is energy-conserving and physically motivated: albedo, roughness, metalness, and a BRDF that behaves like a real surface. Its great virtue is that it responds correctly to any lighting — move a character from a sunlit street into a dim interior and the material still reads as leather, or metal, or skin, with no re-authoring.
NPR throws that out for control. Cel shading, banded lighting, hand-placed rim lights, outline passes, painted-in highlights. Its virtue is that it looks drawn. Its cost is that it tends to be lighting-fragile — an NPR character authored to look good under one lighting setup often looks flat or wrong under another, which is a serious problem in a game with many maps and times of day.
Keeping PBR’s highlights specifically is a smart compromise, because specular response is where the eye reads material and form most strongly. A character can be cel-shaded in its diffuse, keeping the graphic look, while its speculars behave physically — so Cyclops’ visor still catches the light correctly as he turns, and his suit still reads as the material it is supposed to be, under whatever the map is doing.
Destruction without physics
The team built a custom shattering system with automated building cutting tools and debris drop effects that simulates destruction through visual effects rather than physics calculations, explicitly for performance.
This is the correct call and it is the opposite of what most people assume a destruction system is.
Real-time rigid-body destruction is expensive in a way that compounds badly: fracturing a building produces hundreds of bodies, each needing collision detection against each other and the world, and the cost is unpredictable — it spikes exactly when the most is happening on screen. In a competitive shooter, a frame-time spike during a team fight is not a visual problem, it is a fairness problem.
Authored VFX destruction inverts every one of those properties. The automated building cutting tools pre-fracture geometry offline, so the expensive geometric work happens at build time. The debris drop effects are authored particle and mesh animations with a known, fixed cost. The result is deterministic and budgetable — you know what destruction costs before you ship it, and it costs the same every time.
The trade is that it is less reactive. Pre-authored destruction does not respond to how you destroyed something in the way a simulation does. For a game where destruction is environmental flavour and tactical sightline change rather than a physics sandbox, that is a trade worth making — and the team made it deliberately rather than discovering it late.
Cross-platform and VFX legibility
Two more decisions, briefly, because both are about the same underlying constraint:
Asset tiering and strict rendering budget management accommodate different hardware. Which is the unglamorous reality of shipping a game to PC, console and increasingly mobile: not one art pipeline but a family of them, with a defined budget per tier that the art is authored against rather than optimised into afterwards.
VFX uses colour, atmospheric rendering and rhythm manipulation to differentiate each character’s abilities while maintaining clarity in combat. “Rhythm manipulation” is the interesting term there — timing and pacing as a distinguishing property of an effect, not just its colour and shape. In a game where you need to know what just happened to you in a fraction of a second, an ability’s characteristic rhythm is as identifiable as its palette, and it survives being partially occluded.
What to take from it
“Which parts of physical correctness do I keep?” is a better question than “stylised or realistic?” The NPR/PBR split here is a worked example: keep physics where it earns you robustness, stylise where you want control.
Move cost from runtime to build time wherever you can. Pre-fractured geometry and authored effects are the same move as the blendshape and modal-basis work we have written about this week — do the expensive thing once, offline, and leave the runtime cheap and predictable.
And author to a budget rather than optimising into one. Asset tiering before production is much cheaper than a performance pass after it.