Chapter 2: Scan / Kitbash Libraries & CAD Pitfalls

Created by Sarah Choi (prompt writer using ChatGPT)

Scan / kitbash libraries & CAD pitfalls for prop concept artists

Why hybrid matters

Hybrid 2D/3D workflows let you move from a rough idea to a persuasive, production‑ready prop with speed and credibility. Blockouts give you scale and stance. Scans and kitbash parts inject authentic surface language. CAD supplies exacting geometry for hinges, fasteners, and interfacing faces. The danger is that each shortcut can import errors—unit mismatches, topology landmines, licensing traps, and visual noise that breaks gameplay readability. This article treats concept and production artists equally, aiming to help you ship art that stays fast in pre‑vis and stays reliable in downstream pipelines.

Blockouts first: the scaffolding that makes reuse safe

A sturdy blockout is the single greatest accelerant in a hybrid method. Start with an intent‑driven massing pass that communicates proportion, grip/hand clearance, footprint, and primary axes. Color‑code parts by function—structure, covers, fasteners, optics—so later swaps from scans or kitbash don’t scramble intent. Keep edges honest at this stage; use broad bevels that represent manufacturing realities rather than sugar‑cube primitives. In 2D, quick value block‑ins over orthographic guides do the same job: they establish silhouette hierarchy so photo/scan overlays cannot invent form where none exists.

Scans: authenticity with responsibilities

Photogrammetry and LiDAR scans are unbeatable for mid‑frequency detail—cast texture in knurled knobs, the micro‑dings on powder‑coated steel, or the grain of injection‑mold parting lines. To keep scans useful, capture with controlled, diffuse lighting, include a scale reference, and rotate the subject relative to camera to avoid specular bias. The cleanup should target three things: watertightness, decimated but uniform triangle density, and removal of baked‑in lighting. Bake albedo by using cross‑polarized capture when possible or, at minimum, even illumination; otherwise, you’ll fight shadow‑baked textures when relighting. For concepting, a medium‑res, well‑UV’d mesh with a clean normal/roughness set is ideal. For production, ensure naming, alignment to world axes, and a neutral origin so rigging and VFX can instance or simulate without hidden transforms.

Building and curating a scan library

Treat your scan library like a museum with metadata. Every asset should carry capture notes (device, lens, method), scale, real‑world dimensions, material tags, and licensing/consent. Tagging by interface is powerful: “M5 cap screw,” “1913 rail wedge,” “¼‑20 threaded insert,” or “bayonet twist‑lock” tells a designer how parts connect before they even open the file. Versioning is essential. If you remove baked lighting or fix topology, bump the version and preserve the prior one for reference renders. A healthy library is small but precise; fifty impeccable, documented parts outperform five hundred mystery meshes.

Kitbash libraries: speed without trope soup

Kitbashing shines when you need plausible busy‑ness fast. The trap is style‑drift and repetition. Cure it with a house style for edges, radii, and screw logic. If your world uses countersunk Torx and chamfered 1.5 mm edges, normalize third‑party parts to match. When a donor piece is iconic (a GoPro latch, a specific DSLR hot shoe), either alter proportions beyond recognition or explain provenance in‑world. Keep kitbash units consistent—set an asset standard, e.g., meters in DCC and millimeters in CAD—and freeze transforms before export so downstream placement doesn’t inherit hidden scale. For concept artists, use kitbash to block believable joints; for production artists, conform those joints to a briefed interface standard so animation, FX, and UI can count on them.

Photobash ethics and credit hygiene

Photobashing scanned textures or found photography is normal, but ethics define professional practice. Only use sources with licenses that permit derivative works for your production. Keep a source sheet per concept: thumbnails of every external image with links, author, and license. This protects you, credits creators, and helps producers secure rights for marketing splash art. Avoid culturally sensitive objects without context or permission, and do not misuse conservation scans or museum assets restricted to education. If a photobash element dictates a key design choice—a ceremonial motif, a national emblem—document intent so narrative and legal can weigh in early.

CAD in the hybrid stack: precision with sharp edges

CAD is perfect for hard interfaces: hinge knuckles, detent grooves, gear center distances, gasket landings. It is also the number‑one source of pain when exported blindly. NURBS exports with default tessellation often create long, skinny triangles, T‑junctions, and floating normals that explode under subdivision or baking. Always export with explicit chord height and angular tolerances, then inspect density around small fillets; over‑tessellation there kills performance while under‑tessellation there produces faceting in glints. Prefer STEP as an interchange for CAD‑to‑CAD and export triangulated FBX/OBJ for DCC. If your pipeline supports USD, package part hierarchies and materials as USD variants; downstream teams can swap LODs without renaming.

Common CAD pitfalls and how to dodge them

The first pitfall is unit mismatch. An imported STEP in inches dropped into a meters‑based DCC can produce props that are 39.37× off. Lock the studio standard and bake a unit‑check into your import scripts. The second is non‑manifold geometry. CAD loves zero‑thickness surfaces for cosmetic splits; game engines and renderers do not. Before handoff, thicken shells sensibly or boolean‑union cosmetic panels to a parent body, then re‑cut visible seams as normal maps. The third is fillet mania. Real products favor a small set of fillet radii tied to tooling; random radii across a kitbash screams fake. Choose two or three house radii and stick to them. The fourth is material fiction. CAD shaders default to chrome‑like plastics with impossible roughness. Replace them with PBR‑credible materials as soon as geometry stabilizes so lighting and lookdev don’t chase ghosts.

Topology, normals, and baking sanity

When moving from scans/CAD to render‑efficient meshes, prioritize even polygon flow that follows load paths and silhouette. Eliminate ngons and pole clusters at highlight‑critical edges. Triangulate before baking to avoid tangent‑space surprises. Cage bakes solve 80% of normal errors; the rest are usually hard/soft edge mismatches or UV splits at the wrong places. Keep texel density honest across adjacent parts or step it intentionally to guide gaze. For concept‑speed, you can cheat with tiling trim sheets and decals; for production, consolidate UDIMs thoughtfully and plan atlas strategies for props likely to appear together.

Readability and the danger of detail

Scans and kitbash parts carry rich micro‑detail that can drown silhouette. Before polishing, step back to the five‑meter read: can a player identify the class, the affordances, and the safe‑to‑touch areas? Remove medium‑frequency chatter where it muddies the contour. In 2D paintovers, paint broad value fields over the scan to re‑impose graphic order; in 3D, reduce normal strength or reproject micro detail only into the zones that deserve it, like grip panels and wear edges. Production artists should validate readability in engine lighting with gameplay FOV and motion blur, not in a neutral turntable.

Interfaces, tolerances, and animation foresight

Hybrid props often fail at the seams. If a cap is supposed to twist‑lock, show the cam path in orthos and leave clearance. If a service panel hinges, model the knuckle axis and provide stop geometry. Scans rarely encode tolerances; CAD can over‑idealize them. Make a habit of drawing a quick exploded view that labels fastener type, thread direction, and motion range. Animators will bless you, and your kitbash library will become more reusable as a result.

Library governance: keep your speed assets fast

A fast library starts with naming. Use a scheme that encodes category, interface, and scale: “FASTENER_TorxM5x12_cs_v03” beats “bolt_newNEW”. Store preview renders at consistent angles with a checker backdrop and a 10 cm cube for scale. Keep per‑asset readme notes short and present‑tense so another artist can drop the part in minutes. Retire assets aggressively when they cause more fix‑ups than time saved; speed is the KPI. For cross‑team use, maintain a short style guide that defines bevels, screw families, chamfer vs fillet preference, color coding, decal usage, and allowed surface noise.

Photobash overlays on 3D: doing it cleanly

When painting over a blockout or CAD screenshot, separate light and albedo by using blend modes strategically. Treat the 3D render as your lighting plate and add photographic albedo/roughness information on top, not the other way around. Use a neutral, linear‑ish base render and add AO separately so you can dodge/burn for emphasis without crushing detail. If you’re bashing mechanical features like vents or fasteners, align to perspective by eye, then commit to repainting edge highlights and shadow anchors so the photo element truly marries the 3D form.

Handoff notes for downstream teams

Rigging wants pivot accuracy and transform cleanliness; give them zeroed origins, named axes, and sample animation notes for your moving parts. VFX wants named material slots and emissive logic spelled out, plus occluder meshes if your prop glows. UI needs clear status surfaces with screen‑space reserves for indicators. Audio appreciates suggested contact materials (metal‑on‑rubber, ceramic‑on‑metal) and any intentional rattle points. When your hybrid source includes scans, include a note about real‑world materials so Foley and shader authors don’t misread plastic for painted metal.

Red flags checklist

If you see any of the following, stop and fix before you proceed: textures with baked shadows, STEP imports that arrive at comic or microscopic scale, kitbash screws of three unrelated head families in one prop, bevel widths that collapse under game FOV, decal stacks that z‑fight, or fillets that violate your house radii. If the prop only reads after you add grime, the design likely lacks a strong silhouette; fix the form, not just the finish.

A short, practical exercise

Pick a household latch or hinge. Scan it with a phone, clean it to a 100–250k tri mesh, and tag it for your library. Build a neutral CAD version with three house radii and export it with good tessellation. Drop both into your DCC, kitbash them onto a simple door panel blockout, and paint a quick overlay that designates status lights and finger paths. Finish by authoring a one‑page handoff sheet with an exploded view, fastener callouts, and a unit‑checked scale bar. You’ve just created a reusable, ethical, and production‑savvy hybrid asset.

Closing

Hybrid speed is earned, not free. Scans buy texture truth, kitbash buys shorthand, and CAD buys function precision. The cost is diligence: metadata, units, topology, materials, and ethics. When you build your library and habits around those guardrails, your concept art accelerates without surprising the teams who must ship your ideas on screen.