Architect-Ant: Editable Automatic Furnishing of Architectural Floor Plans

Researchers introduce Architect-Ant, an AI system that automatically furnishes architectural floor plans using a fine-tuned vision-language model and a new dataset of 270 professionally designed floor plans. The framework generates furniture layouts as editable symbolic code that can be rendered into realistic images while maintaining spatial validity and functional plausibility.

Architect-Ant addresses a practical gap in architectural technology by combining three key innovations: a curated dataset of real floor plans with furniture annotations, a compact domain-specific language for representing layouts, and constraint-aware AI generation. The dataset itself, AntPlan-270, fills a significant void in machine learning—the lack of professionally annotated furniture arrangement data has hindered progress in this domain. The system's ability to encode architectural constraints like wall alignment, door clearance, and circulation patterns into procedural reasoning traces represents a meaningful advance in spatial AI reasoning.

The framework's dual-representation approach—maintaining both editable symbolic code and photorealistic renderings—addresses a common tension in generative AI. Users can modify generated layouts at the symbolic level without losing semantic meaning, while the Flux-based LoRA renderer produces professional-quality visualizations. This bridges the gap between computational efficiency and visual fidelity. The preference optimization step further refines layout quality beyond initial generation.

For the real estate and architectural technology sectors, this development enables faster interior design prototyping and visualization workflows. Architectural firms and interior designers could significantly reduce manual layout time, while real estate platforms could generate furnished mockups automatically for property listings. The research suggests a scalable path for furnishing large structure-only datasets, indicating potential for widespread application across digitized building archives. The work demonstrates how constraint-aware AI can move beyond surface-level pattern matching toward functionally valid design generation.

• →AntPlan-270 dataset provides the first large-scale collection of professionally annotated furniture arrangements for architectural floor plans
• →Domain-specific language representation enables editable, symbolic furniture layouts that preserve both computational validity and user control
• →Procedural reasoning traces encoding architectural constraints improve model understanding of spatial relationships and design rules
• →The system produces geometrically valid and functionally plausible layouts while rendering photorealistic blueprint-style visualizations
• →Scalable approach could automate furniture arrangement across large structure-only floor plan datasets in real estate applications