Hyper3D Uses AI for 3D Print Prep: From 10M Meshes to Part Split

AI-generated 3D models have progressed rapidly, but generation is only the first step toward a successful 3D print. A digital asset typically needs repair, refinement, fabrication preparation, and often division into multiple printable components before it becomes a finished physical object.

Hyper3D, a generative 3D platform created by Deemos, used Formnext Asia Shenzhen to demonstrate that AI can automate many of those downstream steps. Centered on the Rodin Gen-2.5 generation model, the company presented tools that rebuild messy meshes, edit specific regions of a finished asset with prompts, split complex objects into printable parts, and assign color regions for multi-color printing workflows.

From AI generation to high-resolution geometry

A model that looks detailed on-screen may not contain the physical geometry required for a high-quality print. Surface textures and normal maps can simulate detail visually, but a 3D printer depends entirely on the mesh’s geometry. Rodin Gen-2.5, released in May, provides five geometry-density levels to match different needs. The fastest mode, Extreme-Low, is optimized for rapid ideation, with generation times measured in seconds. At the other extreme, Extreme-High can produce meshes with up to 10 million faces for highly detailed prints. Intermediate settings balance geometry, file size, and processing time: Medium suits figurines, props, and prototypes, while higher-density modes are intended for detailed sculptures, collectibles, and high-resolution resin printing.

More polygons are not always better. Very dense meshes capture fine detail but increase file size and demand more processing during slicing. For many consumer prints, excessive geometric density is unnecessary; what matters is having sufficient geometry where it will impact the final print. Rodin Gen-2.5 gives users the flexibility to choose outputs ranging from lightweight, quick models to highly detailed meshes suitable for premium prints. According to Deemos’ CTO Qixuan Zhang, the model moves “from million-polygon generation in seconds to raw outputs exceeding 10 million polygons,” offering a broad spectrum of geometric fidelity.

Better topology, fewer manual fixes

Topology—the arrangement and connection of polygons—has a strong influence on how easily a model can be edited and prepared for printing. Disorganized triangles, gaps, or overlapping surfaces can cause issues even when a model appears clean in a viewport. Hyper3D’s Smart Low-Poly rebuilds the mesh to follow an object’s structural features and supports both triangle and quad outputs. Cleaner topology eases edits like thickening regions, adding holes, or adjusting areas for supports, and also reduces the computational load on slicers and downstream tools.

Edit a single region without regenerating the whole model

Rarely is a design perfect on the first try. Often only a small part of a model needs revision—a new accessory for a figurine or a structural change to a product component—while the rest is acceptable. Typical AI 3D workflows force users to regenerate the entire model to make any change, losing parts that were already right. Hyper3D’s Local Editing (also called Local Modify or Partial Redo) lets users select a region and modify it with a prompt while preserving the rest of the asset. This enables a productive loop of generate, inspect, edit, refine, and print instead of repeatedly regenerating complete models.

Splitting complex models into printable parts

Large or intricate models often cannot be printed as a single piece because of build volume limits, overhangs, support complexity, and assembly needs. Traditionally, splitting a model requires manual mesh cutting and verification that each part is printable. Hyper3D’s BANG part-separation automates much of this work by analyzing geometry and dividing the model into separate, closed components. The process can be applied recursively, so a character can be split into body, head, and accessories, and those components can be further divided or refined as needed.

After separation, each piece can be independently oriented to maximize build volume use and minimize overhangs. Dividing large objects into sections simplifies post-processing, painting, and final assembly. Parts can be designed to join with magnets, connectors, or other mechanical features. This capability moves AI-generated models beyond visual appeal toward practical design-for-manufacturing workflows.

Preparing models for multi-color printing

As multi-color 3D printing grows more accessible, manually assigning colors in a slicer becomes a time-consuming task, especially on detailed models that require many distinct surface regions. Manual workflows risk misplacing colors, missing regions, or creating boundaries that don’t match the model’s structure. Hyper3D’s color separation analyzes both visual and structural cues to identify coherent regions. Users choose how many color areas they want, preview the result, and export to 3MF, a format that retains color and material data alongside geometry—unlike STL.

Automating color assignment reduces manual effort in multi-color workflows and can improve throughput for production environments and print farms by cutting preparation time.

Is AI becoming a production tool?

End-to-end, the workflow accepts text prompts, single images, or multi-view references and exports to formats such as 3MF, STL, OBJ, FBX, and GLB—covering both 3D printing and broader content pipelines. It can also accept third-party models for repair, retopology, and separation, not just assets created within Hyper3D. Rodin Gen-2.5 is available through integrations like Bambu Lab’s MakerWorld MakerLab image-to-3D workflow, which surfaces the technology to desktop users already preparing files. That sort of integration positions Hyper3D as a processing layer between generation and fabrication, though wider adoption will depend on whether other ecosystems follow suit.

Other companies are developing similar capabilities—automatic part segmentation, watertight outputs, and automatic splitting of oversized models are becoming more common. Hyper3D distinguishes itself by emphasizing post-generation control: its 3D ControlNet lets users constrain proportions and geometry with bounding boxes, voxel grids, or point clouds; Part Separation can be automated or manual; and Local Editing enables targeted edits to selected regions without altering the remainder of the model.

As consumer 3D printers become faster, more affordable, and increasingly capable of multi-color work, the bottleneck shifts from hardware to the availability of usable, printable assets. Traditional 3D modeling still requires training and time, leaving an opportunity for tools that streamline the path from generation to fabrication. The demonstrations in Shenzhen show the workflow is viable; the remaining questions are reliability and consistency—how well these tools handle complex geometry, produce genuinely printable parts, and minimize manual cleanup at scale.

Hyper3D is offering a 14-day free trial of Rodin Gen-2.5; the article noted a promo code of 3DNATIVES for sign-up. If you’ve printed an AI-generated model, consider sharing your experience to help evaluate how effectively these tools perform in real-world production and hobbyist environments.

*Cover Photo Credit: Hyper3D