Every 3D asset that reaches a game goes through roughly the same stages. Knowing them makes it obvious where a generator fits and where it does not.
The stages
- Concept. A drawing, a photo, a reference board. Establishes silhouette and material before anyone models.
- Block-out. Rough volumes at correct scale, dropped into the level to check it reads at gameplay distance.
- High poly. The detailed version — sculpted or hard-surface modelled. Never ships.
- Retopology. A clean, low-density mesh matching the high poly's shape. This is what ships.
- UV unwrapping. Flattening the low poly so textures can be applied.
- Baking. Transferring detail from the high poly into normal, ambient occlusion and curvature maps on the low poly.
- Texturing. Authoring the PBR channels.
- Rigging and skinning, for anything that deforms.
- Export and integration. Correct format, scale, pivot, collision, LODs, materials wired up in engine.
What generation replaces
Mostly concept and high poly. A generator produces a detailed mesh from a description or a photo, which is what a sculpt would have produced — so the stages that consume a high poly still apply unchanged.
That is the important part: generation collapses the front of the pipeline and leaves the back intact. Retopology, UVs, baking and integration are all still there.
What it does not replace
Judgement about the target. Nothing in a generator knows your polygon budget, your texture memory, your engine's material setup, or how close the player gets. Those decisions were always the technical artist's, and they still are.
A shortened pipeline
For background and volume work many teams run: generate → retopologise → unwrap → transfer the generated texture → export. That skips concept, high poly sculpting and hand-authored texturing entirely, and is perfectly adequate for assets nobody inspects closely.