Interior Styling

The technology:

An existing room — photographed or rendered — transformed into different interior design styles using ComfyUI with Flux. Works from a real photograph without needing a 3D model. Multiple style variants generated in parallel for comparison.

Purpose: Decision Support | Design Development Technical Approach: AI-Enhanced Reality: Pure Vision

Complexity: 2 - Intermediate Cost: €€ 3D Model: No 3D Model Required

Interior styling transforms a real photograph or a basic 3D render of a room into multiple interior design interpretations — different furniture styles, material palettes, lighting moods, and spatial character — without rebuilding the room in 3D. The room geometry stays fixed; everything inside it can change.

This is primarily a decision-support tool. It answers the question “what could this space become?” before any design work is commissioned.

How it works

The source is a photograph of the existing room or a basic render from 3ds Max. We run it through a ComfyUI workflow:

Structure preservation — ControlNet extracts depth and structural lines from the source image, anchoring the generation to the room’s actual geometry. Walls stay walls. The window stays where it is. The floor plane stays level. Furniture, materials, and lighting are free to change.

Style generation with Flux — the interior design direction is described as a structured prompt: furniture style period and origin, material palette, lighting character, texture density, atmosphere. Flux generates a photorealistic version of the room in that language, with the ControlNet conditioning ensuring it still reads as the same physical space.

Batch comparison — multiple styles are generated in the same session. A client can compare Japandi, Warm Industrial, French Contemporary, and Maximalist versions of the same room side by side and respond to images rather than vocabulary.

Selective refinement — areas of the image can be masked and refined independently. The overall Scandinavian character is kept; the kitchen joinery is re-run with a different material. The sofa is changed without regenerating the whole room.

Working from a photograph

No 3D model is required. A good-quality photograph of the space — or even a reasonable smartphone image with consistent lighting — is sufficient as source material. This makes interior styling accessible at very early project stages, before any design documentation exists:

A photograph of an empty apartment shell before fit-out
An existing space being considered for renovation
A site visit photograph taken during a client meeting
A real estate photograph of a property being sold

The AI works with what the photograph contains. Low image quality, extreme lens distortion, or very poor lighting will limit result quality.

Style range

The technique handles a wide range of interior traditions:

Scandinavian / Nordic — pale wood, linen, controlled minimalism
Japandi — the Japanese-Scandinavian hybrid: natural materials, negative space, restraint
Warm Industrial — exposed structure, dark metal, warm artificial light
Mid-Century Modern — organic shapes, walnut, statement lighting
Contemporary Luxury — stone, lacquer, integrated lighting, tonal palette
French Apartment — herringbone floors, plaster walls, antique and modern mixed
Maximalist / Eclectic — layered pattern, high color, collected objects

Historical period styles (Art Deco, Victorian, Baroque) are also possible but produce more stylistically illustrative results than the contemporary interior styles above.

What this produces and what it doesn’t

Generated interior images show the character, color palette, material mood, and furniture proportion of a design direction. They do not specify products, produce a furniture schedule, or generate anything dimensionally accurate. The sofa shown might not exist; the flooring pattern might not be a real product.

The value is alignment and direction — confirming that a client and designer are imagining the same kind of space before proceeding to detailed design.

What we need from you

Source image	Photograph or render of the room. Reasonable quality and lighting. Wide-angle or standard lens preferred — extreme fisheye distortion degrades ControlNet conditioning.
What to keep	Any fixed elements that should not change — structural columns, an existing fireplace, a window the client wants preserved. We mask these from the generation.
Style directions	Which styles to explore. References are welcome — a Pinterest board, a hotel lobby, a magazine image. The more specific, the more targeted the generation.
Output use	Client presentation, social media, print? This affects the resolution and number of variants we produce.