Sarah Clarity Studios

Chapter 3: Scan / Kitbash Collaboration & CAD Pitfalls

Created by Sarah Choi (prompt writer using ChatGPT)

Scan / Kitbash Collaboration & CAD Pitfalls for Mecha

2D/3D hybrid workflows are at their strongest when they let you borrow real-world believability without borrowing real-world limitations you didn’t intend. Scans, kitbash libraries, and CAD-derived parts can dramatically speed up mecha design, but they can also create hidden problems: inconsistent scale, impossible assemblies, unusable topology, licensing risk, and “design drift” where your authorship disappears under the source material.

This article is written for both concepting-side and production-side mecha concept artists. If you’re concepting-side, the goal is to gain speed and realism while keeping your design system intact and staying policy-compliant. If you’re production-side, the goal is to receive hybrid work that is structured, cleanly documented, and safe to translate into game-ready assets.

Why scans and CAD show up in mecha workflows

Mecha sits in a sweet spot for scanning and CAD reuse because the subject matter is mechanical. Even stylized mecha often relies on plausible joints, fasteners, vents, hoses, and manufacturing logic. A scan of a real object can give you believable surface breakup and wear patterns. A CAD-derived part can give you crisp precision for pistons, bearings, gear housings, and hard-surface interfaces.

The upside is speed and credibility. The downside is that scans and CAD are not made for concept art by default, and they are almost never made for real-time production. If you import them without a plan, you can end up with a mech that looks “real” but cannot be built, cannot be rigged, and cannot be shipped.

Collaboration reality: you are rarely alone in 3D

In many studios, the concept artist is not the person building the heavy 3D assets. You might be collaborating with a modeler, a tech artist, an outsourcing vendor, or a scan/CAD specialist. The hybrid pipeline works best when you treat that collaboration like a small production: clear asks, clear constraints, and clear handoffs.

For concepting-side artists, the most important habit is to specify what you need 3D to do. Do you need a fast cage to keep proportions consistent? Do you need believable joint assemblies? Do you need a library of “greeble families” that match faction language? If you can name the purpose, you can avoid overbuilding.

For production-side artists, the most important habit is to communicate what is acceptable upstream. If the concept team is using CAD or scans, define early whether those sources can enter the shipping pipeline, whether they must be rebuilt, and what documentation is required.

Scans: what they’re good for (and what they’re not)

Scans are strongest for capturing irregular reality: dents, abrasion patterns, casting seams, rubber texture, fabric weave, and subtle surface waviness that makes things feel manufactured rather than perfectly “CG.” They are also great for reference: scanning a tool, boot, ladder rung, or cockpit component can ground scale and ergonomics.

Scans are weak for clean design control. They often contain noise, uneven density, holes, and messy edges. They may be too literal, locking you into an existing object’s identity. In mecha, that can be a problem if the scan’s silhouette is strong and begins to dominate your design.

A good rule is to use scans as a secondary layer: surface truth and detail inspiration, not primary silhouette authorship. Keep your primary shapes authored by you.

CAD: why it looks amazing and why it breaks everything

CAD geometry can look like a dream: razor-sharp edges, precise circles, perfect fillets, and real engineering believability. The problem is that CAD models are built for manufacturing documentation, not for rendering, animation, or real-time. Common CAD pitfalls include extremely dense tessellation, non-manifold geometry, overlapping parts, tiny hidden fillets that explode polygon counts, and topology that is impossible to UV or bake cleanly.

CAD can also create a design trap: it encourages you to trust the part as “correct” even when it doesn’t fit your mech’s fiction, scale, or style system. A mech designed by stacking CAD parts can feel like a machine catalog, not a character.

The goal is to treat CAD as a reference and a starting point, not as a final asset. Use it to learn shapes and interfaces, then simplify and redesign into your authored language.

The “truth layers” model: how to integrate sources without losing control

A practical way to keep hybrid workflows healthy is to think in layers of truth.

The first layer is the cage or blockout truth: scale, proportions, pivots, and silhouette. This is your authored foundation.

The second layer is the engineering truth: joints, actuators, mounts, recoil supports, and clearances. This is where kitbash and CAD can help—if you keep them subordinate to your cage.

The third layer is the surface truth: panel breaks, fasteners, vents, decals, micro wear. This is where scans and photo textures can add credibility.

If you maintain this hierarchy, your mech remains yours. If you reverse it—starting with scan/CAD surface truth and building upward—you risk ending with an un-authored, unbuildable design.

Kitbash collaboration: building a library that serves style, not speed alone

Kitbash is often treated as a personal shortcut, but the best kitbash systems are shared language. For a team, a good kit library is a style asset: it encodes faction motifs, fastener types, panel rhythms, vent shapes, and actuator silhouettes.

Concepting-side artists should collaborate with production-side artists on what belongs in the kit. If the kit contains parts that are impossible to model cleanly, or that violate performance budgets, it becomes a trap. Production-side artists should push for parts that are modular, scalable, and consistent—parts that can be used across variants without feeling repetitive.

A healthy kit includes both “macro” parts (shoulder housings, knee assemblies, backpack thruster clusters) and “micro” parts (vents, bolts, clamps, cable brackets). The macro parts must be generic enough to support authorship. The micro parts must be consistent enough to unify the world.

Scale mismatch: the quiet killer of scan/CAD workflows

Scale mismatch is one of the most common hidden failures. A scan might be of a handheld tool, but it gets used as a mech ankle component. A CAD bolt might be real-world size, but it becomes a mecha torso detail. The result is a mech that feels wrong even if it looks detailed.

The fix is boring but essential: always bring in a human proxy and a simple scale reference grid, and always test the mech at gameplay camera distance. If your scan/CAD detail disappears at distance, it likely doesn’t deserve texture budget. If your bolts are as big as a pilot’s fist, your scale language is broken.

For concepting-side artists, this is a readability and believability issue. For production-side artists, it directly affects asset budgets and the feasibility of baking and texturing.

Topology and performance: why concept meshes must not pretend to be production

Hybrid workflows blur boundaries. A concept artist can produce something that looks like a model, and a production artist can be tempted to “just use it.” That is where CAD and scan sources become dangerous.

Scan meshes are often triangulated chaos. CAD imports can be millions of polygons with tiny fillets. Neither is suitable as a shipping asset in most real-time pipelines. Even if you decimate them, you often lose clean edge flow and bake quality.

The best practice is to treat scan/CAD geometry as reference or proxy, not as a production mesh, unless the studio has a specialized pipeline for remeshing and rebuilding. In handoff, label your files clearly: “proxy,” “reference,” “for paintover,” or “not for shipping.” Production teams should set policy that concept meshes are not assumed to be usable unless they meet defined criteria.

The “CAD fillet trap” and other shape problems

CAD models often contain fillets and chamfers that are physically correct but visually wrong for your design. In games, you may need exaggerated bevels to catch highlights at distance. CAD fillets can be too small, producing razor edges that look dead under typical lighting.

CAD also encourages perfect symmetry and perfect circles. Mecha can look sterile if everything is perfect. You may need deliberate asymmetry, heroic exaggeration, and simplified plane breaks that read clearly. Treat CAD precision as a reference, then re-author the shape language so it reads.

Another pitfall is the “part stacking” look: too many nested brackets, plates, and bolts that feel like a real machine but obscure the silhouette. If your mech reads as a pile of parts rather than a designed character, you’ve lost your hierarchy.

Clearance and animation: CAD realism vs rig reality

CAD assemblies often ignore the kinds of clearance problems that show up in character animation. Armor overlaps might be fine in a static CAD pose but intersect horribly in a run cycle. A piston might be “correct” but placed where a rig cannot maintain it through motion.

This is where the cage blockout remains the boss. Even if you use CAD joints, you must test range-of-motion and collision zones. For concepting-side artists, do quick extreme poses and mark collision risk areas. For production-side artists, push back early if the concept implies impossible articulation, and propose redesigns before the asset is built.

Photobash ethics in scan/CAD heavy workflows

When scans and CAD are involved, the ethical and legal boundaries become sharper, not softer. A scan of an object you do not have rights to may be problematic, especially if it captures distinctive industrial design. CAD models downloaded from the internet often have unclear licensing, and some are copyrighted or restricted.

The safest standard is to use studio-provided scan libraries, your own scans, or properly licensed assets. Avoid scanning branded products or importing CAD of recognizable real-world items if the design could be considered protected. In many studios, the policy is: reference is fine; direct reuse is not, unless licensed.

For concepting-side artists, document your sources. If you used a scan pack, note whether it is internal and approved. If you used external CAD, verify licensing. For production-side artists, do not assume concept-source assets are legally safe for shipping; treat them as reference unless cleared.

Photobash texture ethics follow the same logic. Use licensed textures or your own photos. Avoid pulling images from random websites. If the concept must become production reference, keep a clean layered file so any unshippable textures can be removed.

Collaboration checklist: what to ask for and what to deliver

In scan/kitbash collaboration, clarity beats talent. A good request to a 3D collaborator specifies: target scale class, intended gameplay camera distance, articulation needs, key silhouettes, and what passes you need (clay, AO, ID masks, turntable frames). It also specifies what you do not need—so time is not wasted on details that will be painted over.

A good delivery back to the team includes: screenshots of key views, a simple legend for material IDs, notes on what geometry is proxy/reference, and a list of any external sources and their licensing status. If you changed kit parts heavily, note what was altered so others know what is reliable.

For production-side teams receiving this, the most useful additions are: pivot markers, clearance notes, and warnings about any CAD/scan elements that must be rebuilt.

Preventing “kitbash soup”: preserving authorship and style systems

The fastest way to ruin a design with kitbash and CAD is to accept everything as-is. You end up with mixed design languages: different bolt styles, inconsistent panel rhythms, and clashing edge philosophies.

Counter this with a style system. Decide early what your fasteners look like, how thick plates are, what chamfer scale is, how vents are shaped, and how decals are placed. Then “normalize” kitbash parts to those rules. Even a simple step like re-scaling fillets, removing random bolts, and regrouping panels into larger families can make kitbash work feel authored.

Concepting-side artists should do silhouette checks regularly: if you can’t recognize the mech in a black shape, you’re building detail at the expense of identity. Production-side artists should watch for kit repetition: if the same shoulder assembly appears across multiple mechs, the world begins to feel copy-pasted.

Handoff guidance: how to keep scan/CAD helpful without becoming toxic

For concepting-side handoff, include both the pretty paintover and the truth passes: neutral lighting render, AO if available, material ID pass, and at least one “exploded” or annotated view showing part groupings. If CAD or scans were used, explicitly label them as reference/proxy unless the pipeline supports reuse.

For production-side handoff, request that any scan/CAD heavy concepts include a “rebuild intention” note: which shapes are essential to preserve, and which can be simplified. This prevents wasted time trying to replicate CAD micro-detail that will never read.

A strong handoff respects everyone’s time. It gives production clear intent, and it avoids dumping unstructured CAD/scan clutter onto downstream teams.

The core principle: borrow truth, not liability

Scans and CAD can make your mecha feel real, but they can also import liability: legal risk, unmanageable topology, and an un-authored look. The healthy hybrid mindset is to borrow the kind of truth that serves your design—plane logic, interface believability, manufacturing cues—while rejecting the parts that harm readability, pipeline feasibility, or ownership.

If you keep the cage as your foundation, use kitbash and CAD as subordinate engineering helpers, and use scans and photo textures as controlled surface seasoning, you get the best of both worlds. Your mecha will look credible, iterate quickly, and hand off cleanly—without falling into the common scan/CAD pitfalls that slow teams down and erode design clarity.