This week, I used Opus 5.5 + Tripo + Blender to turn the interstellar journey from "The Little Prince" into a 3D game that users can explore personally. After release, it accumulated nearly 3 million views across the web.
https://x.com/Saccc_c/status/2109962425512734739
Many friends loved this game. From scene construction and character creation to code development and debugging, the main production work was handled by AI.
So, I decided to share the production process, combining real cases and referenceable prompts, to guide you step-by-step into AI 3D game creation, allowing you to start building your imagined worlds too.
This article is divided into three parts:
- Game development workflow and preparation
- Scene construction
- Character construction
1. Game Development Workflow and Preparation
Game development workflow: Fable 5.1 handles PRD creation, organizing core scenes, characters, and storylines + Image 2.5 generates scene and character images + Tripo generates core 3D assets + Opus 5.5 builds core code + Astra iterates optimization and bug fixes.
Game tool configuration:
- Blender: Download Blender CLI to optimize Tripo-generated 3D models and complete overall 3D modeling; install the paid Auto-Rig Pro plugin for character rigging and skinning, facilitating animation interaction creation.
- Unity: Install Unity App (https://unity.com/) as the game engine to package scenes, characters, and interactive effects into a complete game; install Unity CLI and add the Pipeline package in the project so Claude can directly operate Unity projects via official Unity MCP.
Next, this article will focus on explaining scene construction and character creation.
2. Scene Construction
Simply put: Image 2.5 generates scene images — decompose 3D assets and generate them in Tripo — Opus 5.5 operates Blender+Unity for scene modeling.
Detailed steps are as follows 👇:
(1) Use GPT Image 2.5 to generate scene reference images, turning ideas into clear visuals.
Based on the previously organized game requirements, use Image 2.5 to generate scene reference images, determining the overall art style, color scheme, and positions of main objects.
It is recommended to first determine a satisfactory scene effect image, then supplement other angles around it (especially for indoor and outdoor scenes). The top-down view mainly illustrates scene layout, while the oblique side view supplements object height, shape, and hierarchical relationships. You can also directly generate an overall view like below:
Reusable reference prompt:
Based on the uploaded scene reference image, help me generate a comprehensive reference image of this scene, including multi-view perspectives of the overall environment and displays of local details.
(2) Let AI decompose scene assets, filtering out models that need to be generated separately in Tripo.
With reference images ready, you can directly use AI to judge which assets in the scene need to be generated separately in Tripo and which can be built by Opus 5.5 in Blender/Unity.
Reference prompt:
I am now proceeding with game development, my tech stack is Tripo + Blender + Unity. Based on the scene reference images we generated above, help me deeply analyze which assets are suitable for generation with Tripo, and which assets are suitable for direct construction with Blender. Mark the assets that need to be generated by Tripo in the image for my confirmation.
For the production of scene prop 3D assets in Tripo, it is recommended to use the H3.1 model, which ensures high-precision details of the assets.
(3) Use professional prompts to let Opus 5.5 control Blender and Unity to start modeling. The key to modeling is providing detailed reference images and iterating from multiple camera angles.
After generating and exporting the 3D props that need separate production in Tripo, hand these assets and the scene reference images from step one to Opus 5.5, letting it control Blender + Unity to build the overall environment, place props according to reference images, and adjust proportions and positions.
After completion, check the scene from different angles such as front and side, combining screenshot feedback to gradually adjust layout and details.
Reference prompt is as follows:
Based on the uploaded multi-view reference images of interior and exterior scenes and the generated Tripo 3D assets, you perform a perfect-level replication against the reference images. Requirements include achieving the same stunning texture and detail restoration as the reference images. After implementation, dispatch an independent review agent to conduct blind audits of the scene from multiple camera positions. If differences from the reference images are found, continuously iterate and improve.
3. Character Modeling
Image 2.5 generates character images — Generate 3D character models in Tripo — Rig and skin using Blender's Auto-Rig Pro plugin — Integrate Mixamo animations — Secondary adjustments for complex interactions in Blender (optional).
Here we take the design of the Little Prince as an example for specific explanation.
(1) Use Image 2.5 to generate multi-view reference images of the character.
Based on the character image and settings, first use Image 2.5 to generate the Little Prince's T-pose four-view diagram, ensuring uniform character details across all views.
(2) Use Tripo's Smart Mesh P2.0 model to generate 3D character assets.
Smart Mesh P2.0 is a 3D generation model specifically designed for games and interactive content. Its feature is generating controllable polygon count and editable character structures, facilitating later rigging and animation production.
In Tripo, select "Multi-view to 3D Generation" and upload the four views sequentially. Default topology is quadrilateral, precision can be set to maximum, select the P2.0 model and start generation.
Afterward, if the initially generated white mesh grid is neat and error-free, click "Texture" to generate model textures; if the mesh shape is unreasonable, choose "edit mesh" for partial regeneration.
(3) Use the Auto-Rig Pro plugin in Blender to automatically rig and skin the character.
After configuring the plugin, directly upload the character model to Claude and input the following prompt, which I have repeatedly debugged and summarized, for efficient rigging and skinning 👇:
Use Auto-Rig Pro to complete rigging and skinning for the character, ensuring bone positions align with character joints. After completion, test basic actions like raising hands, bending legs, and turning around, checking and correcting obvious deformations, pulling, and clipping issues.
(4) Download required action files for the character from Mixamo and let Opus 5.5 adapt them to the corresponding character skeleton.
According to game settings, go to mixamo (https://www.mixamo.com/), search for actions needed by the Little Prince by keywords, and download the corresponding FBX action files.
Afterward, we can let Opus check whether related actions match the Little Prince's actual skeleton and complete integration.
(5) Adjust special actions and contact details in Blender (required for complex interactive actions).
For complex interactions involving character-prop contact like "Little Prince carrying a kettle," Mixamo often does not have perfectly adapted actions. We can first select a close base action, then let Opus 5.5 iteratively adjust details in Blender.
Prompt reference is as follows:
Please refer to the provided Mixamo action, retain applicable parts, and use Auto-Rig Pro's arm IK controllers to adjust arm poses, coordinating finger grip and kettle position detail corrections to complete the kettle-carrying action. Ensure stable palm-handle contact, natural lifting process, avoiding clipping, grip detachment, or floating kettles.
Next week, I will open-source the entire game code and more complete production details on GitHub. Everyone is welcome to stay tuned!
ps. If you also want to try your own 3D creation with Tripo, you can register via my link to get 500 credits for free: https://studio.tripo3d.ai/home?invite_code=H2DB1X&utm_source=X&utm_medium=kol&utm_campaign=kol_Saccc
