How AI 3D Generation Extends What You Can Do with Mobile 3D Scans

Two Tools. Two Different Jobs.

For many creators today, 3D creation often starts from one of two places: an idea, or the real world.

The first is AI generation. Tools like Tripo let you turn a text prompt, a sketch, or a reference image into a 3D model in seconds. It's fast, creative, and requires no scanning equipment or technical setup.

The second is mobile 3D scanning. KIRI Engine, for example, captures objects and environments that already exist in the real world, preserving their accurate real-world geometry, surface texture, and material detail, using nothing but a smartphone.

These aren't competing approaches. They solve different problems. Together, they cover more of the creative pipeline than either tool can handle alone.

What Mobile 3D Scanning Does Well

Mobile 3D scanning captures what already exists: the real shape of a physical object, its actual surface texture, the way light interacts with its materials. This is something AI generation can approximate, but not fully replace when real-world accuracy matters, because the information comes from reality itself.

KIRI Engine offers three capture modes depending on what you're scanning:

• Photogrammetry captures dozens to hundreds of photos and reconstructs accurate geometry and color maps. Best for objects with clear surface detail and texture.
• LiDAR uses the depth sensor on supported iPhones for fast spatial scanning with real-world scale, especially for rooms and larger spaces. In KIRI Engine, LiDAR scans can also be enhanced with AI to improve surface detail, making them more useful as spatial references or scene foundations.
• 3D Gaussian Splatting (3DGS) captures the visual appearance of objects and scenes without starting from traditional mesh geometry. This can be useful for complex materials, organic surfaces, and full environments where traditional photogrammetry may be harder to reconstruct cleanly. KIRI Engine also includes built-in 3DGS-to-Mesh conversion, allowing creators to turn splat captures into mesh assets on mobile when they need polygonal models for downstream workflows.

The result is a high-fidelity 3D capture that serves as a strong foundation for further creative work.

What AI Generation Adds to a Scanning Workflow

High-fidelity scans provide a strong real-world foundation. For different downstream uses, creators may still need lighter meshes, new material directions, or additional generated assets to complete a scene. This is where Tripo extends the workflow:

• Retopology reduces polygon count significantly while keeping the scanned geometry visually recognizable, producing clean Quad topology suitable for real-time workflows and 3D tools.
• Texture generation replaces or transforms the surface material of a scanned asset, opening up entirely new visual directions from the same geometry.
• Asset generation creates new 3D objects and characters that can be placed directly into scanned real-world environments.

Each of these expands what creators can do with a scan after capture.

Workflow 1: From Real-World Sculpture to Game-Friendly Asset

Real example: a painted figurine scanned with KIRI Engine

A detailed figurine makes a strong test case for this workflow. The original piece has complex surface detail across clothing folds, facial features, and a dynamic mid-stride pose. Exactly the kind of geometry that would take hours to model manually but captures well with photogrammetry.

Step 1: Scan with KIRI Engine

Using Photo Scan mode, capture the object from multiple angles and heights. For a figurine of this size, 60 to 80 photos covering the full silhouette is typically enough. KIRI Engine processes the reconstruction automatically and exports a GLB file with geometry and texture maps intact.

Step 2: Restyle with Texture Generation

In this test, we found that running texture generation before retopology produced better results. With the full scan geometry intact, Tripo has more surface information to work with when interpreting the material direction. Reducing polygon count first may limit some of the surface information available for texture generation.

With that in mind, using the prompt "stylized fantasy RPG villager, hand-painted game texture, warm earthy colors, leather tunic, rough fabric clothing, worn sandals, painterly details, clean material separation, low-poly game asset style", the scanned figurine was reinterpreted as a stylized fantasy RPG-style character asset. The pose and proportions came from the real-world scan. The visual style came from Tripo.

Step 3: Retopologize in Tripo

With the visual direction established, the asset is then retopologized to reduce polygon count. The original scan came in at 40,296 triangles. After retopology, the same asset came out at 9,566 faces with Quad topology, roughly a 76% reduction, while keeping the pose, proportions, and overall silhouette intact.

One physical object. One scan. In this workflow, a real-world sculpture can become a lighter, stylized game-friendly asset without manually rebuilding it from scratch.

Workflow 2: When a Real Person and an AI Character Share the Same Scene

Real example: a 3DGS-captured person combined with an AI-generated character

One of the more compelling uses of 3D Gaussian Splatting is capturing real people. 3DGS is particularly well-suited for capturing a person together with their surrounding environment in a single continuous pass, preserving realistic lighting, clothing detail, and spatial context all at once.

Step 1: Capture a real person with KIRI Engine 3D Gaussian Splatting

Scanning a person with 3D Gaussian Splatting requires a bit more care than scanning an object. The subject needs to stay as still as possible during capture, even small movements like breathing or shifting weight can introduce blur in the reconstruction. We asked the subject to hold a relaxed pose and minimize movement throughout.

Capture from three height levels: high angle, eye level, and low angle, walking around the subject at each height. This gives the algorithm enough information to reconstruct the full figure accurately. KIRI Engine processes the footage into a 3DGS scene that includes both the person and the surrounding environment, preserving the real lighting and spatial context of the location.

Step 2: Generate a character with Tripo

For more control over the final result, we used Tripo's text-to-image generation first to create a reference image, then fed that image into Tripo's image-to-3D workflow. While text-to-3D is useful for fast ideation, using a generated reference image first gave us more control over the final visual direction.

The resulting character was exported as a standard 3D mesh. Its visual style, with flat shading, clean lines, and saturated colors, contrasts intentionally with the photorealistic scanned figure.

Step 3: Combine in Blender using KIRI Engine's 3DGS plugin

3DGS scenes are exported as .ply files. To bring them into Blender, we used KIRI Engine's open-source Blender plugin, which loads the 3DGS scene as a native object alongside standard 3D assets. The Tripo-generated character was imported as a standard mesh and positioned in the scene alongside the scanned figure.

KIRI Engine’s Blender plugin loads the real-world 3DGS capture as a scene object, while the Tripo-generated character is imported as a standard 3D mesh. Together, they create a mixed real-and-AI scene inside the same Blender workspace.

The result is a single scene where a real person captured through 3DGS and an AI-generated stylized character coexist in the same space. From here, the scene can be used as a starting point for several creative directions: game prototyping, concept visualization, virtual production tests, or mixed real-and-AI storytelling.

The scanned elements provide authenticity and spatial grounding. The AI-generated character adds content that does not exist in the physical world. Together, they show how a real-world 3DGS capture can become the foundation for a broader creative scene, rather than remaining a static scan.

Scanning Gives You the Real World. AI Generation Gives You Everything Beyond It.

Neither tool replaces the other. Tripo is the right choice when you're building something from imagination, when the asset doesn't exist yet and speed and creative flexibility matter most. KIRI Engine is the right choice when accuracy to reality matters, when the object or environment already exists and needs to be captured rather than invented.

Used together, the two tools cover the full range of 3D asset creation: from pure imagination to pure reality, and everything in between.

Try the workflow yourself: capture something real with KIRI Engine, generate or adapt creative assets with Tripo, and see how reality-based capture and AI generation can work together.