Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DreamHome-Pano: Design-Aware and Conflict-Free Panoramic Interior Generation

Lulu Chen, Yijiang Hu, Yuanqing Liu, Yulong Li, Yue Yang

TL;DR

DreamHome-Pano addresses the conflict between rigid interior layouts and flexible stylistic guidance in 360-degree panoramic generation. It introduces a Design-Aware Prompt-LLM to translate layouts and style references into coherent, geometry-compliant prompts, and a Conflict-Free Control framework to decouple structure from appearance via an empty-room normal map and standardized reference templates. The training employs a hierarchical, progressive pipeline (SFT in three stages and two-stage RL with DPO and Diffusion-NFT) on a large, curated panoramic dataset plus expert-curated aesthetics, yielding professional-grade results demonstrated on a standardized benchmark of 50 layouts and 10 styles and outperforming baselines like Seedream 4.5 and Gemini 3 Pro Image. This work advances AI-driven interior visualization by ensuring spatial fidelity and design plausibility, with practical impact for architectural visualization and design workflows.

Abstract

In modern interior design, the generation of personalized spaces frequently necessitates a delicate balance between rigid architectural structural constraints and specific stylistic preferences. However, existing multi-condition generative frameworks often struggle to harmonize these inputs, leading to "condition conflicts" where stylistic attributes inadvertently compromise the geometric precision of the layout. To address this challenge, we present DreamHome-Pano, a controllable panoramic generation framework designed for high-fidelity interior synthesis. Our approach introduces a Prompt-LLM that serves as a semantic bridge, effectively translating layout constraints and style references into professional descriptive prompts to achieve precise cross-modal alignment. To safeguard architectural integrity during the generative process, we develop a Conflict-Free Control architecture that incorporates structural-aware geometric priors and a multi-condition decoupling strategy, effectively suppressing stylistic interference from eroding the spatial layout. Furthermore, we establish a comprehensive panoramic interior benchmark alongside a multi-stage training pipeline, encompassing progressive Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). Experimental results demonstrate that DreamHome-Pano achieves a superior balance between aesthetic quality and structural consistency, offering a robust and professional-grade solution for panoramic interior visualization.

DreamHome-Pano: Design-Aware and Conflict-Free Panoramic Interior Generation

TL;DR

DreamHome-Pano addresses the conflict between rigid interior layouts and flexible stylistic guidance in 360-degree panoramic generation. It introduces a Design-Aware Prompt-LLM to translate layouts and style references into coherent, geometry-compliant prompts, and a Conflict-Free Control framework to decouple structure from appearance via an empty-room normal map and standardized reference templates. The training employs a hierarchical, progressive pipeline (SFT in three stages and two-stage RL with DPO and Diffusion-NFT) on a large, curated panoramic dataset plus expert-curated aesthetics, yielding professional-grade results demonstrated on a standardized benchmark of 50 layouts and 10 styles and outperforming baselines like Seedream 4.5 and Gemini 3 Pro Image. This work advances AI-driven interior visualization by ensuring spatial fidelity and design plausibility, with practical impact for architectural visualization and design workflows.

Abstract

In modern interior design, the generation of personalized spaces frequently necessitates a delicate balance between rigid architectural structural constraints and specific stylistic preferences. However, existing multi-condition generative frameworks often struggle to harmonize these inputs, leading to "condition conflicts" where stylistic attributes inadvertently compromise the geometric precision of the layout. To address this challenge, we present DreamHome-Pano, a controllable panoramic generation framework designed for high-fidelity interior synthesis. Our approach introduces a Prompt-LLM that serves as a semantic bridge, effectively translating layout constraints and style references into professional descriptive prompts to achieve precise cross-modal alignment. To safeguard architectural integrity during the generative process, we develop a Conflict-Free Control architecture that incorporates structural-aware geometric priors and a multi-condition decoupling strategy, effectively suppressing stylistic interference from eroding the spatial layout. Furthermore, we establish a comprehensive panoramic interior benchmark alongside a multi-stage training pipeline, encompassing progressive Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). Experimental results demonstrate that DreamHome-Pano achieves a superior balance between aesthetic quality and structural consistency, offering a robust and professional-grade solution for panoramic interior visualization.
Paper Structure (42 sections, 4 equations, 17 figures, 5 tables)

This paper contains 42 sections, 4 equations, 17 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Panoramic Image Generation
  4. Controllable Image Generation
  5. Data
  6. Hierarchical Curation Pipeline
  7. Stage 1: Massive Foundational Dataset
  8. Stage 2: Quality-Filtered and Diversity-Sampled Dataset
  9. Stage 3: Expert-Curated Aesthetic Dataset
  10. Hierarchical Attribute-Grounded Captioning
  11. Step 1: Element and Attribute Extraction
  12. Step 2: Fluent Caption Composition
  13. Method
  14. Prompt-LLM
  15. Training Data and Style Prior
  16. ...and 27 more sections

Figures (17)

  • Figure 1: Evaluation of DreamHome-Pano. (a) Automatic evaluation; (b) Human evaluation.
  • Figure 2: Showcase of panoramic generation guided by floorplans and style reference images.
  • Figure 3: Showcase of the multi-style generation capability of DreamHome-Pano.
  • Figure 4: Hierarchical curation of high-quality panoramic interior images: quality filtering, clustering-based sampling, and expert aesthetic refinement.
  • Figure 5: Hierarchical Attribute-Grounded Captioning Pipeline.
  • ...and 12 more figures