Textures, UVs & Real-World Scale

A texture is a flat 2D image, but your model is 3D, so something has to decide which part of the image lands on which part of the surface. That something is the UV map. Every vertex of a mesh carries, besides its 3D position, a 2D coordinate - labelled U and V to avoid clashing with X, Y, Z - that says 'this point of the model corresponds to this point of the texture image'. Laying out those coordinates is called UV unwrapping: conceptually, flattening the 3D surface out onto the 2D texture like peeling and pressing flat the skin of an orange.

Good UVs matter because they control texture size, direction and continuity. If a wall's UVs are stretched, the texture stretches with them - bricks smear. If two adjacent faces have mismatched UV scale, the brick size jumps at the seam. Most architectural models arriving through Datasmith (Module 2) bring UVs from Revit, Rhino or SketchUp, and they are often serviceable but rarely ideal - which is why understanding and, where needed, fixing UVs is a core texturing skill.

A crucial distinction for archviz: many surfaces use tileable textures repeated across large UVs rather than a unique unwrap. A brick wall does not need every brick individually painted; it needs a small seamless brick texture repeated. So the UV job for most walls and floors is not a delicate unwrap but simply setting the right repetition and scale so the tile reads at life size. Unique unwraps are reserved for things that genuinely need a one-off texture, like a specific signage panel or a hero object.

Almost all architectural texturing runs on tileable (seamless) textures - images whose edges match so they repeat endlessly with no visible seam. Brick, plaster, concrete, wood flooring, carpet, gravel: a small seamless tile, repeated. The whole game is then how often it repeats, and that is where real-world scale comes in.

Real-world scale means the texture repeats at its true physical size: a brick texture tiled so each brick is roughly 215mm long, floorboards at a believable 150-200mm width, a 600mm floor tile that is actually 600mm. Get this right and the surface reads as real without the viewer knowing why; get it wrong and it looks off even if they cannot name the fault. The reliable way to achieve it is to author or import textures with a known real-world size and set tiling from the surface's real dimensions - a 4m wall showing a 1m-wide texture repeats it four times.

Unreal gives you a powerful shortcut here: World Aligned Texture (and World Position-based tiling) projects a texture in world space at a fixed real size regardless of the model's UVs. Drop it on any surface and a brick is brick-sized automatically, seams and messy imported UVs be damned - superb for quickly getting consistent scale across a whole imported building. The trade is that world-aligned projection can smear on surfaces facing awkward directions and gives less fine control, so it is a fast baseline rather than a universal answer. Many archviz masters expose a 'use world-aligned' switch so you can choose per instance.

Texel density is the single most useful concept for texture quality, and most beginners have never heard of it. It measures how many texture pixels (texels) fall on a given real-world distance - commonly texels per centimetre, or pixels per metre. It is, in effect, the resolution of a surface as seen in the world.

The goal is consistent texel density across surfaces the viewer sees together. If your floor has a high texel density (crisp, detailed) but the adjacent wall has a low one (soft, blurry), the mismatch reads as wrong even though each surface in isolation is fine. Uneven texel density is a classic giveaway of a rushed scene - one wall razor-sharp, the next mushy. Studios often pick a target density (say, a certain number of pixels per metre appropriate to how close the camera gets) and texture everything to roughly match it.

The practical levers are texture resolution and tiling. A large surface can keep good texel density either with a higher-resolution texture or by tiling a smaller one more often - tiling is usually the cheaper answer, which is another reason tileable textures dominate archviz. Surfaces the camera gets close to (a worktop, a door handle you inspect in VR) want higher density; distant context buildings can drop to very low density with no one noticing. Spending your texel budget where the eye actually goes is the whole art. Unreal has an optimisation view mode that visualises texel density and texture streaming so you can spot the mismatches instead of guessing.

For the repeating linear details of a building - skirtings, cornices, window frames, panel mouldings, railings, door trims - there is a smarter approach than a unique texture each: the trim sheet. A trim sheet is a single texture laid out as horizontal strips, each strip a different profile or material band (a moulding here, a metal edge there, a panel groove below). You then lay a model's UVs onto the strip you want, so many different pieces of geometry all draw from one shared texture.

The payoff is efficiency and consistency at once. Dozens of trims, frames and mouldings across a whole building share one texture in memory and one material, which is dramatically cheaper than a bespoke texture per element, and they all match in style and texel density automatically. It is a technique borrowed from game environment art, and it is ideal for architecture precisely because buildings are full of repeated linear profiles.

Trim sheets pair naturally with tileable textures: use tileables for the big flat surfaces (walls, floors, glazing) and trim sheets for the linear detail that frames them. Together they let you texture an entire building convincingly from a small, tightly-managed set of textures - which is exactly what a real-time scene, always fighting a memory and performance budget, needs. The upfront effort of laying UVs onto a trim sheet pays back every time you reuse it, and a good trim sheet becomes a studio asset used across many projects.

Every texture you load occupies GPU memory (VRAM), and a real-time scene has a finite budget - tighter in VR, where you are rendering at high resolution twice a frame and cannot afford stutters from over-full memory. So texture resolution is a constant negotiation, not a 'bigger is better' choice.

The honest rule: match resolution to how the surface is actually seen. A 4K (4096px) texture on a worktop the camera leans over is justified; the same 4K map on a ceiling glimpsed once from below is waste. A wall textured with a tiling 2K brick at correct real-world scale can look sharper than a stretched unique 4K map, for a quarter of the memory - tiling beats brute resolution more often than beginners expect. Unreal helps by streaming textures (loading higher mips only as needed) and by generating mipmaps (smaller pre-filtered versions used at distance, which also stop distant textures shimmering), but sensible authored resolutions still matter because streaming manages a budget you set.

A few reliable habits: use tileable textures and trim sheets to reuse a small set of maps; reserve high resolution for hero surfaces the camera inspects; drop context and distant geometry to low resolution; and check the texture-streaming and texel-density view modes to find where memory is being spent badly. Manage this well and a large, detailed building stays comfortably interactive; ignore it and the scene either runs out of memory or grinds - the exact failure VR punishes hardest. Module 9 goes deeper on optimisation; scale and texture discipline set here is where it starts.