Styly Muse
Styly Muse: our generation system and the research behind it
Redesigning a room from a photo is not one problem but several: keeping walls and windows where they are, removing furniture without warping the room, furnishing an empty space believably. Styly Muse handles each with a dedicated workflow.
What Styly Muse is
Styly Muse is our generation system. Rather than one setting applied to every request, it brings together separate workflows: staging an empty room, redecorating a furnished one, removing furniture, visualising floors and walls, placing a product in a room, animating a still. Each workflow has its own instructions and settings, and Muse picks the model suited to the task.
Muse has been refined with feedback from more than 200 professionals — interior designers, decorators and real-estate agents — who used it on their own projects. Their feedback decided what must stay untouched in a room and what is allowed to change.
It builds on published research in AI interior design, which we follow and draw lessons from. The work that matters most to us is listed below, by problem.
01
Keep the room, change the design
What the research established
The central difficulty of redesigning from a photo is preserving geometry: walls, openings, perspective. Research on conditional control of diffusion models, and later on instruction-guided image editing, showed how to change what is in an image while respecting its structure.
In Muse: the decoration and staging workflows start from your photo and keep the room’s architecture.
Papers
- Adding Conditional Control to Text-to-Image Diffusion Models — Zhang et al., 2023 · arXiv:2302.05543
- LooseControl: Lifting ControlNet for Generalized Depth Conditioning — Bhat et al., 2023 · arXiv:2312.03079
- InstructPix2Pix: Learning to Follow Image Editing Instructions — Brooks et al., 2022 · arXiv:2211.09800
- FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space — Black Forest Labs et al., 2025 · arXiv:2506.15742
- Qwen-Image Technical Report — Wu et al., 2025 · arXiv:2508.02324
02
Generation built for interiors
What the research established
Several teams have shown that a general-purpose model is not enough for interior design: it takes coherent styles, furniture at the right scale and plausible materials. These papers describe models and systems specialised for the domain.
In Muse: 18 design styles and separate settings per room type.
Papers
- RoomDiffusion: A Specialized Diffusion Model in the Interior Design Industry — Wang et al., 2024 · arXiv:2409.03198
- iDesigner: A High-Resolution and Complex-Prompt Following Text-to-Image Diffusion Model for Interior Design — Gan et al., 2023 · arXiv:2312.04326
- DiffDesign: Controllable Diffusion with Meta Prior for Efficient Interior Design Generation — Yang et al., 2024 · arXiv:2411.16301
- VIDES: Virtual Interior Design via Natural Language and Visual Guidance — Le et al., 2023 · arXiv:2308.13795
- Personalized Interiors at Scale: Leveraging AI for Efficient and Customizable Design Solutions — Zhou et al., 2024 · arXiv:2405.19188
03
Removing furniture without damaging the room
What the research established
Erasing large objects from a room easily produces inconsistencies: a wall that bends, a floor that changes. Research on furniture removal shows how to use the room’s layout to rebuild what was behind the furniture correctly.
In Muse: the furniture-removal workflow, which empties a room before it is refurnished.
Papers
- Layout Aware Inpainting for Automated Furniture Removal in Indoor Scenes — Kulshreshtha et al., 2022 · arXiv:2210.15796
- An Empty Room is All We Want: Automatic Defurnishing of Indoor Panoramas — Slavcheva et al., 2024 · arXiv:2405.03682
04
Staging and light
What the research established
Furnishing an empty room believably means respecting its light: shadows, reflections, direct sun. These papers decompose the appearance of an empty room so that objects can be inserted consistently and the scene relit.
In Muse: the home-staging workflow for property listings.
Papers
- Semantically Supervised Appearance Decomposition for Virtual Staging from a Single Panorama — Zhi et al., 2022 · arXiv:2205.13150
- Digital Kitchen Remodeling: Editing and Relighting Intricate Indoor Scenes from a Single Panorama — Ji et al., 2025 · arXiv:2504.16086
05
Working with designers
What the research established
Recent research studies co-design between designers and AI: how a generative tool fits into a professional’s work rather than replacing it.
In Muse: a system refined with more than 200 professionals, and the ability to edit a result rather than regenerate everything.
Papers
- AIDED: Augmenting Interior Design with Human Experience Data for Designer-AI Co-Design — Lin et al., 2026 · arXiv:2602.10054
06
Foundations
What the research established
The work that made high-resolution image generation by diffusion possible, and on which most of the field rests.
Papers
- High-Resolution Image Synthesis with Latent Diffusion Models — Rombach et al., 2021 · arXiv:2112.10752
These publications belong to their authors; we cite them as the work Muse builds on, without claiming to reproduce them exactly or to speak for their authors.