Hyper3D's Rodin Gen-2 brings conversational edits to 3D models

Anyone who has experimented with AI 3D generators knows the real headache isn't making an asset, it's tweaking it. The moment a headlight looks slightly off or a carving feels wrong, your only real option was hitting reroll and praying the next entire model turned out better.

Hyper3D is changing that dynamic with Rodin Gen-2. The headline feature is a new targeted editing tool that brings inpainting workflows straight into three-dimensional space.

Instead of tearing down a model, creators can drag a bounding box over a specific region and tell the AI what to change with plain text. In hands-on testing, generating a vintage wooden chair with ornate carvings was just step one. By isolating the chair back with a box selection and prompting "add a floral pattern to the carvings," the model updated the selected spot in seconds while leaving the rest intact.

Natural language meets mesh editing

The 3D Inpainting feature handles both geometry tweaks and material swaps. In another test using an image of a classic car, the system produced the base vehicle, then accepted follow-up prompts to apply a metallic finish to the paint and reshape the headlights.

Behind the curtain, Rodin Gen-2 combines a large-scale diffusion model with a specialized 3D representation. It runs a masked diffusion process over the targeted coordinates, seamlessly blending the replacement geometry into the untouched areas. Spatial guidance can also be anchored using bounding boxes, voxels, and point clouds, alongside an automated smart low-poly optimizer.

Crucially, the output avoids the messy, unworkable triangulation typical of older generative tools. Rodin Gen-2 outputs production-ready quad meshes with ready-to-use UV maps. The assets export cleanly into STL for 3D printing, alongside standard formats like FBX, OBJ, GLB/GLTF, and USDZ, with direct pipeline integrations for Blender, Unity, and Unreal Engine.

Why it matters

We see partial editing as the tipping point for AI 3D assets moving from concept art into actual production pipelines.

In 2D image generation, tools didn't truly fit professional workflows until inpainting arrived. 3D modeling is exponentially more complex and expensive to rework manually. By letting creators box-select a problem area and fix it with a conversational prompt, Hyper3D turns an unpredictable slot machine into an interactive creative partner. Indie game developers and 3D printing hobbyists can now salvage near-perfect assets instead of starting from scratch.

This move puts serious pressure on competing platforms like Tripo3D and Scenario to step up their localized control systems. Hyper3D's decision to focus on clean quad topology and ready-to-print STL exports is especially smart, directly targeting the friction points artists complain about most.

That said, natural language still has limits. Text prompts can tweak a texture or morph an ornament, but they won't replace parametric precision when millimeter-accurate engineering tolerances are on the line. For concept props, quick game prototypes, and custom prints, however, it cuts hours of tedious cleanup down to seconds.