Photogrammetry to Blender: Cleanup, Retopo, and Export

A photogrammetry export is usually a high-poly triangle mesh with a baked color texture. If you captured it with KIRI Engine's Photo Scan, that raw export can look great, but it often still needs cleanup, retopo, UV checks, and texture baking before it becomes a usable Blender asset.

Where You're Starting From

Before anything else, confirm what you're holding: a dense triangle mesh (often hundreds of thousands to millions of polygons depending on your photo count and object complexity), one baked color texture, and no UV layout beyond what supports that one texture.

If you have a Pro plan, you have a decision to make before you even open Blender.

Decision Point: KIRI's Quad-Mesh Retopology, or Manual Retopo in Blender?

KIRI Engine's Quad-Mesh Retopology (Pro only) converts the default triangular mesh into a quad-mesh automatically on export, with UV maps unwrapped as part of the conversion. Triangle meshes are efficient for reconstruction and rendering, but they are harder to manage in workflows that need regular edge flow, clean subdivision, predictable deformation, or controlled UV editing. Quad meshes are easier to adjust, subdivide, and deform, which is why they are common in game development, film production, and asset workflows that require further editing.

For most props, environment pieces, and static assets, using this automatically is the right call. You lose fine-grained control over topology flow, but you save hours.

Manual retopo in Blender gives you full control over edge flow, which matters if you're rigging the asset for deformation (a character, a creature, anything that needs to bend) or if you need topology optimized for a specific LOD target.

Practical rule: if it's a static prop or environment piece, use Quad-Mesh Retopology and move on. If it needs to deform, retopo by hand, or at minimum plan to clean up the auto-retopo result around joints and deformation zones.

A Photo Scan export usually starts as a dense triangle mesh. Quad-Mesh Retopology converts it into a cleaner quad-based mesh that is easier to edit, subdivide, and prepare for downstream workflows.

Step 1: Import and First Checks

Import the mesh (OBJ, FBX, or GLTF, depending on what you exported). Before doing anything else, check three things:

Scale. Photogrammetry scans don't inherently know real-world scale unless you gave your scanning tool a reference. Check your mesh dimensions against the real object. If it's off, fix scale now, before you do any other work, because it affects modifiers, physics, and export later. If you're using KIRI Engine, you can measure and correct scale before export using the app's built-in Measuring and Scaling tool, rather than eyeballing it after the fact in Blender.

Normals. Recalculate normals outside (Blender: Mesh > Normals > Recalculate Outside). Photogrammetry meshes occasionally generate inverted normals in areas with poor photo coverage.

Origin point. Set the origin to something sensible (geometry center or base of the object) before you start editing. It's a two-second fix now versus a headache later if you're exporting to a game engine.

Step 2: Clean Up Geometry

Look for floating geometry (disconnected islands of mesh that photogrammetry sometimes generates from background elements or reflections) and delete them. Run Merge by Distance (Mesh > Clean Up > Merge by Distance) to collapse duplicate vertices left over from reconstruction, then check for non-manifold geometry (Select > Select All by Trait > Non-Manifold), which shows up often in photogrammetry meshes at seams between photo clusters and can cause problems later in retopo or baking.

Look for holes, most commonly on the bottom of the object if you didn't flip it during capture, or on the object's back if you didn't get full angle coverage. If you scanned with KIRI Engine's Photo Scan and used the turntable capture method or scanned the bottom separately, you should have fewer holes to deal with here.

For small holes, Blender's Fill and Grid Fill tools handle most cases. For larger missing sections, you're better off rescanning that angle than trying to sculpt a fix that will look wrong under the baked texture.

If the file is too heavy to navigate comfortably, use a temporary Decimate modifier to work with a lighter version while you inspect the mesh. Decimation is a viewport and performance aid at this stage, not a substitute for retopology; a decimated triangle mesh is still a triangle mesh with irregular, unpredictable topology, just with fewer of them.

Step 3: Retopology (If You're Doing It Manually)

If you're retopologizing by hand: use Blender's Poly Build tool or a shrinkwrap-based retopo workflow (build a low-poly cage, shrinkwrap it to the high-poly surface). Focus edge flow around areas that will deform or that need silhouette accuracy. Everywhere else, keep the topology as light as your quality target allows.

If you scanned with KIRI Engine and used Quad-Mesh Retopology, this step is done for you, with UVs already unwrapped as part of the conversion. Check them for stretching around high-curvature areas, then move to baking.

Step 4: UV Unwrapping (Manual Path Only)

If you retopo'd manually, mark seams along natural edges (hidden seams where possible) and unwrap. Check your UV islands for overlap and stretching in Blender's UV editor before baking anything.

Step 5: Baking Textures

Before baking, make sure your high-poly scan mesh and your retopo'd low-poly mesh are aligned in the same position and scale in the scene; misalignment here is the most common cause of a bad or noisy bake. Keep both objects: the high-poly scan as your detail source, the low-poly retopo as the mesh you'll actually bake onto and use going forward.

If you're on KIRI Engine's Pro plan, this is where PBR Material Generation can reduce manual material work. KIRI Engine's PBR pipeline uses a diffusion-based AI model, the technology behind modern AI image generation, to reconstruct material properties such as albedo, roughness, and surface detail, instead of relying only on a single baked color texture. In practice, this gives you a fuller material set built for PBR workflows, where color, roughness, and reflectivity need to be handled separately in Blender, Unreal, or Unity.

KIRI Engine’s PBR Material Generation helps turn a Photo Scan export into a fuller material set, separating color and surface properties such as roughness and reflectivity for Blender, Unreal, and Unity workflows.

If you're baking manually: with your aligned high-poly and low-poly meshes in the same scene, use Blender's bake panel to transfer detail from high to low poly. Common maps to bake:

This is the standard game-asset baking workflow and works the same way regardless of whether your source mesh came from photogrammetry or manual sculpting.

Step 6: Common Problems

Texture looks stretched or blurry in spots. Usually a UV stretching issue. Go back to the UV editor and check island distortion in that area.

Mesh looks "melted" or has soft, undefined edges. This is a capture problem, not a Blender problem. It usually means the object had shiny, reflective, or low-texture surfaces that standard photogrammetry struggles with. If you're using KIRI Engine, use Featureless Object Scan for that object type instead.

Scale is wrong after import. Check your unit settings in both your scanning tool's export and Blender's import settings. Mismatched units (meters vs centimeters) is the most common cause.

File size is too large for a smooth Blender session. If you're on KIRI Engine, export settings let you choose between low, medium, high, and ultra poly counts, or set a specific target count on Pro. Ultra is useful when you need maximum surface detail, but for a Blender working file, medium or high is usually enough headroom before you retopo down further.

Bake looks noisy or has dark artifacts. Almost always a high-poly/low-poly alignment issue from Step 5. Double-check both meshes share the same origin, scale, and position before re-baking.

A Professional Reference Point

A footwear studio comparing KIRI Engine against Reality Capture used Quad-Mesh Retopology with a 50% target quad count and 8K textures, reporting a game-ready result without manual cleanup. Full breakdown, including the exact capture setup: Fast Photogrammetry: KIRI Engine vs Reality Capture for a White Vegan Shoe

If You're Building for a Game Engine

Everything above gets you a clean, editable Blender asset. If your endpoint is Unity or Unreal specifically, there's an additional layer: LOD generation, collision meshes, and engine-specific texture packing. That's covered in Photo Scan for Game Assets: Unity and Unreal Workflows.

Scanning with a Different Method?

Everything above assumes you're working with a mesh. If you captured with 3D Gaussian Splatting instead, you're working with a point cloud of splats, not a triangle mesh, and the Blender workflow is completely different. See KIRI's 3DGS Blender Add-on for viewing and rendering splats directly in Blender.