Built this playable 3D game with GPT-6 Astra + Hyper3D MCP.
GPT handled the game design and coding, while Rodin generated the detailed 3D models and game assets directly through MCP.
No manual coding or modeling — basically one prompt → 3D assets → a game you can actually play.
Kinda crazy how simple this workflow is getting.
I’ve been testing GPT-6 Astra + Hyper3D MCP for 3D workflows lately, and this combo actually makes a lot of sense.
GPT-6 Astra is already pretty good at using Blender for structured geometry like buildings, vehicles, and other hard-surface assets. But with characters, creatures, or more organic shapes, building everything directly in Blender can sometimes drift from the reference.
What worked better for me was:
➡️ GPT-6 Astra analyzes the reference and plans the workflow
➡️ Hyper3D Rodin MCP handles the character or creature generation
➡️ If I need joints, connectors, assembly parts, or specific pieces separated, I ask GPT to use BANG to split the model where needed
➡️ Then I bring everything into Blender for rigging, animation, editing, and rendering
Spent about half an hour drawing this little cutie and decided I had to turn it into something real.
Used Hyper3D Rodin to turn the drawing into a 3D model, then used BANG to Parts for the part separation. It split the model into separate printable pieces on the first try, and the separation was surprisingly clean. I only needed a little cleanup before printing.
The textured version was already ridiculously cute, and honestly, almost every angle looked like it could be a profile pic.
Since I was going to paint it myself anyway, I went ahead with the print. This tiny thing only took about an hour to print, and after adding the colors, it turned out even cuter.
Pretty satisfying seeing a quick drawing turn into a tiny physical figure.
Tried turning a concept image into a full 3D scene using Hyper3D Rodin.
Instead of generating the whole scene at once, I broke it into individual assets including the food truck, sign, robot, monitors, barrels, and smaller props.
I used Extreme-High for the main geometry, Micro for smaller details like scratches and surface wear, then Native 12K PBR textures for the materials.
The 12K textures were probably the biggest improvement for me. When I pushed the camera closer, chipped paint, stickers, printed graphics, text, and logos still stayed surprisingly sharp. I also checked the edges, grooves, and backs from different angles, and the texture coverage held up really well without obvious stretching or blurry spots. Even the wear and differences between metal and plastic surfaces came through nicely.
After that, I brought everything into Blender for composition, material tweaks, lighting, and the final camera work.
Pretty efficient workflow for turning a 2D concept into a detailed 3D scene.
Wanted to see how well 12K PBR textures actually hold up in close-ups, so I picked a T-Rex skull as the first stress test.
Generated it with Hyper3D Rodin using Extreme High + Micro, then textured it with Extreme-High + Native 12K.
It already looked pretty good from a normal distance, but zooming in was the interesting part. The cracks, weathering, mineral deposits, and tiny details around the teeth were still surprisingly clear.
This AI-powered animatronic Mangle was made by @BioMakers Industries using Hyper3D Rodin, then printed with Bambu Studio.
The model was prepared for 3D printing using BANG to Parts. It can automatically separate a model into components, while Manual mode lets you choose exactly what to split — like the head, hands, or different color areas. Individual parts can also be BANGed again for further decomposition.
The parts were then exported as STL / 3MF, sliced and printed with Bambu Studio, assembled, fitted with motors, and programmed. The jaw actually moves.
Compared with a traditional workflow, where each component and color section would need to be modeled and separated manually, BANG to Parts cuts down a lot of the manual work, especially for part separation and color-based splitting.
The car was created using Hyper3D Rodin, then brought into Codex to build the actual game.
The BANG feature was really useful for breaking the car into individual parts without losing quality. I could split individual parts again for further decomposition, then reassemble everything for more iterations.
It was pretty handy to go from AI-generated 3D assets to game-ready parts.
This model was generated with Hyper3D Rodin and then turned into an actual 3D print.
I used Extreme-High for this one, which can push the model up to 10M+ polygons and capture some really fine, skin-texture-level details.
The scales, horns, spikes, and all those weird little details came through surprisingly well in the final print. I honestly wasn't sure how much of that detail would actually survive the printing process, but seeing it as a physical object was pretty crazy.
Just played around with Rodin Extreme‑High mode.
With solid reference inputs, it hits skin‑texture‑level detail and can go up to 10M+ polygons.
The leather grain on this bag turned out incredibly crisp. It’s honestly mind‑blowing how far image‑to‑3D tech has come lately.
Used Rodin Gen-2.5 to turn my drawing into a 3D model. I went with the new Micro mode and 12K textures.
The 12K texture quality was probably my favorite part. With full PBR materials and no blind spots, even the smaller clothing details came through surprisingly well. At this point, making little AI-generated figures feels almost too easy.
Tested by T-Boy on Rodin Gen-2.5 update, using its extreme-high & high modes.
Hyper3D Rodin is the first 3D AI to achieve 10M+ polygons, the micro-level surface details are incredible!