Autodesk

What LookdevX is

LookdevX is Autodesk's look-development tool: where artists build what a surface is made of, brushed brass, wet asphalt, skin, as a graph of connected nodes. It is built on open standards, USD and MaterialX, so a material authored once can travel between Maya and the other tools in a studio's pipeline.

My work on it was about the start: making a powerful, abstract tool approachable for artists opening it for the first time.

Problem statement

LookdevX needed a lower learning curve, and its fastest-growing users, students on education licences, were meeting node-based shading for the first time with nothing visual to hold onto.

Who was actually arriving

The assumed user was a seasoned look-development artist. The data said otherwise.

Working with Autodesk's analytics team, I found that a large share of new users were arriving through universities and education licences. That reframed the brief. We had been trying to reduce LookdevX's learning curve for professionals; the people growing fastest were meeting node-based shading for the first time.

The same data explained where to invest. Many long-time Maya users were still in Hypershade, Maya's older material editor, out of habit. New users had no habit. They were adopting LookdevX, the cross-compatible tool, far more quickly, which made it the place where a better start would pay off most.

I paired the numbers with what users said. I ran surveys on Reddit and read through the internal forum where beta testers reported on the platform. The request that kept coming back was thumbnails: people wanted to see what each node was doing, so they could manipulate the graph visually and watch the result change in real time, rather than reading parameter values and imagining the material.

A node graph is abstract until you can see what flows through it. For someone new to shading, that abstraction is the learning curve.

01Show the material on the node

I designed thumbnails for the nodes themselves, so every surface in the graph carries a live preview of what it produces, and a material viewer that previews the result on a sphere or a cube under different lighting without leaving the graph. The feedback loop moved from change a value, render, compare, to change a value and watch.

For experienced artists it saves renders. For new ones it does more: it teaches the graph, because cause and effect are finally on the same screen.

02Start from a material, not from zero

A default surface node opens as a column of numbers. For a student, the fastest way to learn what those numbers do is to start from something recognisable and change it. So I designed templated material presets, chrome, car paint, velvet, skin, glass, honey and more, that you choose by looking rather than by name.

My first design rendered a large preview of each material as you hovered its name in the list.

Engineering reviewed it and told us it couldn't be built as designed. So I kept the intent and changed the form. Each preset got a small thumbnail sitting next to its name in the list, visible all at once rather than one at a time on hover. The list became scannable at a glance, which is arguably better for someone choosing among thirty materials.

The next question was what happens once you edit one. A preset you have changed is no longer that preset, but it isn't nothing either. When any value moves away from the preset's defaults, the label becomes Custom (from Gold), so it is honest about the change and still says where it came from. The last customised version stays in the list, so you can try another preset and come back to your own work. It remains temporary until you save it deliberately from the contextual menu.

Exploration is only safe if you can get back. The custom state is what lets a beginner try things without fear of losing the thing they liked.

Also at Autodesk: an early generative 3D character concept

Late in my time at Autodesk I was brought in to work out, in full and in two weeks, what generating 3D characters could look like. It was a small team: a product manager, another product designer focused on a different feature, and weekly reviews.

The core idea: a prompt alone is too blunt to direct. So the builder pairs a written description with structured choices, age group, height, build, skin tone and visual style, lets you start from references, and offers variations to pick from rather than a single roll of the dice. Once a character exists, you refine it by region, face, clothing, or a selected accessory, instead of regenerating the whole thing and losing what already worked.

WATCH THE CONCEPT · 1:04

The direction continued after I left, led by other designers, and evolved into what is now Wonder 3D. What's shown here is the early concept, not the shipped product.

What I would do differently

My first presets design, with large hover previews, couldn't be built. Pairing with an engineer in week one would have reached the thumbnail approach sooner and saved a design round.

Reflection

The most useful research finding was not about the interface. It was about who was arriving. Knowing that new users came through education changed what "reduce the learning curve" meant, from streamlining for experts to making the graph legible to people seeing one for the first time.