How AI interior design comes together: from Blender to real-time exploration
Thomas Ricouard’s house project connects requirements, floor plans, Blender/Python, Cycles renders and UE5 through repeated checks and revisions.
From the campus into a room
After our campus experiment, we looked at how AI might help create an interior. Thomas Ricouard shared a house made with Astra, moving from a Blender scene to a standalone Unreal Engine 5 application that could be explored in real time.
This is an analysis of his published case, not a reproduction we completed ourselves. The useful part is understanding the work between the initial request and the explorable result.
1. Describe the space before asking for a plan
The creator specified room counts and a modern, minimalist house in a quiet setting with trees. Astra produced a floor plan. The lesson is to describe spatial requirements rather than simply asking for a more sophisticated result.
For our own project, we would also specify area, fixed doors and windows, room uses and furniture requirements, then check room connections. These are our suggestions, not the creator’s complete prompt.
2. Make an editable scene in Blender
The creator shared an early modeling time-lapse and explained that Astra tended to work through headless Blender or the command line, writing Python rather than clicking through the interface.
He explicitly said that this V1 modeling happened before the floor plan shown above. Requirements → plan → modeling is therefore a helpful explanation, not his exact timeline. The important deliverable is an editable scene that can be revised, rendered and exported.
3. Work on lighting and rendered views
The creator asked for several 4K Blender Cycles renders with a cinematic golden-hour direction and emphasized the importance of lighting.
A convincing interior depends on lighting and composition as well as geometry. Selected views can check furniture, light and materials before moving on. Here, 4K refers to the requested still renders, not every published video.
4. Import into UE5 and walk through
According to the creator, Astra wrote the import process, launched repeated tests and compared the result before delivering a standalone application. A further reply described walking around the house at 60 FPS on his own computer and recording the exploration.
Rendered views show selected compositions; a real-time application lets a viewer choose where to go. His reported frame rate is a result on his setup, not a performance promise for other computers.
5. Add another pass of detail and interaction
The creator repeated the Blender-to-UE5 process, adding props and adjusting lighting to make the house feel lived in. He also said the coffee machine worked.
Our takeaway is to establish a complete path before selecting details to improve. For the campus, this could mean entering one room, seeing the furniture and reaching a window or balcony before expanding the experience.
6. Compare the outputs
The creator asked Astra for a storyboard comparing Blender and Unreal Engine views. Matching views help check whether the intended scene survives the move between tools.
He attributed much of the sky, lighting, camera and video work to Astra. This remains his description: the public material does not disclose all prompts, total token cost, the complete project or every asset source. It cannot establish that every asset was generated from scratch or guarantee one-shot reproduction.
What this process teaches us
A useful outline is requirements → floor plan → 3D scene → lighting and renders → real-time exploration → detail → comparison. Actual work can move back and forth between these stages.
This is an explorable 3D interior proposal that can support discussion of layout and atmosphere. The case does not establish construction-ready renovation delivery.
Start with one question about your room
This analysis draws on the main post and eight publicly visible creator replies checked for the source article. It does not claim to cover all login-only replies or independently reproduce the workflow.
Would a desk leave enough room to move around a small bedroom? Would extra storage darken a living room? A specific room and question give still renders and exploration a shared purpose.
Key Takeaways
- This is a useful process model, not a strict chronology of the creator’s work.
- Still renders and real-time exploration need separate checks.
- The creator’s reported 60 FPS is not a guarantee for every device.
Sources & further reading
Related Articles
Once AI builds a scene, how do you turn it into a playable game?
Emm Tee’s PaperRoute shows the work beyond a generated scene: establish a gameplay loop, then review art, characters and movement separately.
Read MoreA small-group lesson with PikaTo: classroom points, homework checks and feedback
A sample lesson workflow for preparing rules, recording participation, checking homework and reviewing feedback before sharing it with parents.
Read MoreBuild a consistent lesson-feedback routine for your teaching center
Agree on the information every feedback note needs, then use classroom and homework records to keep each student’s feedback specific.
Read More