CG Generalist

What they do

A generalist does the whole 3D chain on a shot rather than one department's slice of many shots: blockout modelling, unwrapping and texturing, shading, lighting, a camera match, a simple simulation if the shot needs one, rendering, and often the comp as well. The role exists in two places. In small facilities, commercials and design work it is the normal way to make a shot, because splitting a two-week job across six specialists costs more than it saves. In large facilities it appears on work that does not justify a full pipeline: a set extension, a hero prop, an insert, a graphics element. The day is decision-dense and tool-wide: a session might involve a modelling package, a sculpting tool, a texture painter, a renderer and a compositor, all before lunch. The generalist's real skill is judging sufficiency, because they cannot afford to do anything to a specialist's standard. A prop seen for eight frames in soft focus needs geometry that reads in silhouette and nothing else. A hero object in a slow push needs proper topology, real texture detail and physically plausible shading. Deciding which is which, correctly, on every element in a shot, is the job. Interchange matters: assets often have to hand off cleanly to a larger pipeline, so scene description, naming and colour management have to be right.

What they decide

The calls that belong to this role and no other.

• How much quality each element in a shot needs, which is a budgeting decision made continuously
• Which parts of the shot are 3D and which are cheaper in 2D
• Modelling approach, topology and level of detail per asset
• Shading and lighting approach, and the renderer and sampling settings that get an acceptable frame in the time available
• How assets are structured and named so they can be handed to a larger pipeline
• When a shot needs a specialist and cannot be finished generalist (raises it; supervisor decides)

Principles of the craft

• Build to the frame, not to the asset. The camera decides how much detail exists. Modelling the back of an object the audience never sees is the most common way a generalist loses a day.
• Silhouette first, surface second, detail last. A shape that reads wrong in outline cannot be rescued by texture. Get the proportions and the outline right at blockout, when changing them is free.
• Reference is the fastest route to plausibility. Measured dimensions, a photograph of the real material and an HDRI of a similar environment eliminate most of the guessing that generalists spend time on.
• Physically plausible shading is a shortcut, not a purity test. Energy-conserving materials with sensible roughness behave predictably under any lighting, which means one lighting change does not require retuning every surface.
• Render time is a design constraint. Sampling, ray depth and resolution decisions are creative decisions on a deadline, because a look that cannot render overnight is not available to you.
• Interchange discipline pays for itself the first time a shot grows. Consistent naming, a portable scene description and a managed colour space are what let a generalist's asset survive being handed to a real pipeline.

Common mistakes

• Modelling and texturing everything to hero standard, then running out of time for lighting, which is what actually sells the shot
• Working without measured or photographic reference and tuning by eye, then failing to match a plate
• Building shading in an unmanaged colour space, so textures and renders do not match the compositor's working space
• Neglecting the comp, when most of the integration and half the perceived quality lives there
• The experienced generalist's mistake: reusing a personal rig of tools and looks so every job resembles the last one
• No scene organisation, so an asset cannot be handed over or revisited three months later

How to learn it

Exercises and routes in, not a reading list.

• Take one photograph of a real object with a ruler in frame and rebuild it: model, texture, shade, light into an HDRI, render, and comp it back into the original photograph. Then show both to someone and ask which is which.
• Do the same shot twice with a two-hour and a two-day limit. Write down what you cut in the short version and whether anyone noticed. That list is your sufficiency judgement, made explicit.
• Light one untextured grey model six ways, changing only the lighting. Compare against six reference photographs. Lighting delivers more perceived quality per hour than modelling, and this proves it to you.
• Export an asset as USD or Alembic, open it in a different application, and render it there. Everything that broke is a pipeline lesson you now own.
• Build a set extension for a real plate you shot: match the camera, build only what the frame needs, and integrate it. Then move the camera and find out what you failed to build.