My Photogrammetry Cleanup Pipeline: From Scan to Sale-Ready Asset

Best 3D Model Platforms

After years of selling 3D assets, I’ve refined a cleanup pipeline that transforms messy photogrammetry scans into professional, sale-ready models. This process isn't just about fixing geometry; it's a philosophy of balancing automation with artistic control to maximize quality and efficiency. I'll walk you through my step-by-step workflow, explain where I lean on manual work versus AI assistance, and share how I package assets for marketplace success. This guide is for 3D artists and scan enthusiasts who want to monetize their work without getting bogged down in technical debt.

Key takeaways:

• Raw scans are data, not assets; they require intentional repair, retopology, and optimization to be usable.
• A hybrid approach, using AI for heavy lifting and manual tools for fine control, yields the best quality-to-effort ratio.
• Marketplace success depends as much on technical preparation (clean topology, logical UVs) as on presentation and accurate pricing.
• Integrating AI-powered preprocessing into my initial assessment phase saves days of manual cleanup on complex organic scans.

Why Raw Scans Aren't Ready for Sale: My Core Principles

The Common Flaws in Raw Data

In my experience, every raw photogrammetry scan comes with inherent issues. The mesh is typically a dense, non-manifold "triangle soup" with millions of polygons, holes from occluded areas, and floating debris from the capture environment. The textures are often baked onto this messy topology, leading to stretching, seams, and lighting artifacts. Selling this raw data is a disservice to buyers—it's unusable in real-time engines and a nightmare for any downstream animation or modification.

My Non-Negotiable Quality Checklist

Before any model leaves my hands, it must pass my checklist. The mesh must be watertight (no holes), manifold (clean edges), and have a logical polygon flow. UVs need to be efficiently packed with minimal stretching. Textures must be clean, artifact-free, and provided in standard PBR sets (Albedo, Normal, Roughness, etc.). Finally, the entire asset must be optimized for its intended use-case polygon count.

How AI Tools Like Tripo Fit Into My Philosophy

My philosophy is to let the machine handle the tedious, computational heavy lifting so I can focus on creative direction and final polish. This is where AI tools become indispensable. For instance, I often use Tripo as a powerful first pass in my pipeline. I'll feed a problematic raw scan into it, and its AI-driven reconstruction can output a clean, base watertight mesh with sensible topology in seconds. This gives me a perfect starting block—it solves the initial data repair and decimation challenge, allowing me to skip straight to the refinement stage. It fits my workflow not as a magic "finish" button, but as a highly intelligent preprocessing assistant.

My Step-by-Step Cleanup Workflow in Practice

Step 1: Initial Assessment & Decimation

My first step is always to audit the scan in a viewer like Blender or Unreal. I look for major holes, scale, and the overall polygon density. My immediate goal is to reduce the insane poly count from millions to a more manageable few hundred thousand without visibly losing shape. I use a decimation modifier, but I'm careful—aggressive decimation on messy scans can create more problems. Sometimes, I'll run the scan through an AI reconstruction tool first to get a cleaner, lower-poly base mesh to work from, which makes this step far more predictable.

Step 2: My Retopology & Mesh Repair Process

This is the core of the cleanup. I need a new, clean topology over the scan data. For hard-surface objects, I often do this manually or with semi-automatic guides. For complex organic shapes (like statues or rocks), I rely on AI-assisted retopology. Using the cleaned mesh from my initial AI preprocessing, I can then use QuadriFlow or similar algorithms to generate a clean, animation-ready quad mesh. I then manually fix any awkward loops, ensure edge flow supports deformation if needed, and verify the mesh is perfectly manifold.

My quick manifold check:

• Run a "Select Non-Manifold" operation.
• Any selected edges/verts must be investigated and repaired.
• Ensure all normals are consistently oriented.

Step 3: UV Unwrapping & Texture Baking

With a clean, low-poly mesh, I now create new UVs. I use smart UV project or manual seams for optimal space usage. The magic happens in baking: I project the high-detail geometry and color information from the original scan onto my new low-poly mesh and its clean UVs. This transfers all the visual fidelity onto efficient, seamless texture maps.

My baking checklist:

• Bake Normals (from high-poly to low-poly)
• Bake Ambient Occlusion
• Bake Curvature (for smart material masks)
• Crucially: Bake the original phototexture as the new Albedo map.

Step 4: Final Polish & Optimization

I open the baked textures in Photoshop or Substance Painter to clean up artifacts, remove shadows or light probes, and enhance details. I then create the full PBR material set. Finally, I optimize the scene: ensure the model is at real-world scale, pivot point is logical, and the final polygon count is appropriate. I often create Level of Detail (LOD) variants for game-ready assets.

Comparing Methods: Manual vs. AI-Assisted Cleanup

When I Choose Manual Refinement

I go fully manual for assets where precision is paramount—architectural elements, product designs, or any asset that requires specific, controlled edge loops. Manual retopology is also my choice when preparing a model for complex deformation, like a character face, where I need absolute control over topology flow for proper rigging and animation.

Where AI-Powered Tools Save Me Days

AI is my go-to for the initial "data sanitation" of complex organic scans. Cleaning a dense, hole-ridden scan of a tree stump or a gothic gargoyle manually is a week-long chore. Using an AI tool to instantly generate a watertight, base-retopologized mesh can reduce that to an hour of setup and refinement. It's not about replacing my skill, but about eliminating the soul-crushing, repetitive parts of the job.

My Hybrid Approach for Best Results

My standard pipeline is hybrid. I use AI for the heavy initial lift: taking the raw, messy scan data and turning it into a clean, workable foundation. This gives me a "Stage 1" asset. I then take that into my traditional software (Blender, ZBrush, Substance) for manual artistry: refining the topology in key areas, perfecting the UVs, painting out texture errors, and setting up materials. This combines the speed of AI with the quality control of hands-on expertise.

Prepping Your Asset for Marketplace Success

My Packaging & Presentation Standards

Presentation is everything. I always provide:

• Rendered beauty shots on a neutral background and in a simple scene.
• A wireframe overlay render to showcase clean topology.
• A texture map preview sheet.
• The final files: FBX/GLTF, clean low-poly mesh, and all texture maps in a standard resolution (4K or 2K).
• A clear README.txt with scale, units, polygon count, and engine-specific notes.

Setting the Right Price Based on Effort

I price based on uniqueness, complexity, and cleanup effort. A simple, cleanly scanned rock might be $15. A complex, fully cleaned, rigged, and textured character asset with multiple LODs can be $200+. I factor in the time my hybrid pipeline saved me—if AI cut 8 hours of manual cleanup, I price for the value of the final asset, not the reduced labor time, but it makes mid-complexity assets much more profitable.

What Buyers Really Look For (From My Experience)

Buyers, especially game developers, prioritize usability. They want assets that drop into their project without work. This means:

• Clean, efficient topology that won't cause rendering issues.
• Logical, non-overlapping UVs with good texel density.
• PBR textures that follow standard naming conventions (e.g., _Albedo, _Normal).
• Accurate metadata—if you say it's 10k polys, it must be 10k polys.
• Modularity where applicable. Show how the asset can be used or combined.