I Went to Google HQ to Learn How to Properly Use AI to Visualize My Home Renovations — Here Are the Tricks I Learned

An architect specializing in AI image generation helped me understand better how to use Google Gemini as a tool in the design process

an ai created room with a rust colored conversation put in an apartment block
(Image credit: Google Gemini / Nano Banana Pro)

Let me say this from the very start — I'm an AI sceptic. Yet, that's almost why I wanted to take up an invitation from Google to spend a morning workshopping with Amir Hossein Noori, an architectural designer who also specializes in using AI in high-end architecture and interior design. I guess I wanted the answer to the question: Is using AI really going to help anyone design a better home? Burying my head in the sand doesn't mean this tool doesn't exist, and that people aren't going to use it.

The goal of the session was to learn how to better use Gemini, Google's AI assistant, and its Nano Banana Pro image creation tool. My cohort for this beginner workshop included several interior designers, so I was curious to know where they felt this tool fit into their businesses. As it turns out, before the age of AI, those three-dimensional renders of projects, used to help clients visualize their finished space based on a designer's scheme, would cost somewhere in the region of £2,000 — so not often viable for smaller design studios and clients with more limited budgets. It's likely, I'd suggest, that AI renders will be the norm for interior design studios in the next year, though most of the designers in attendance were already using them to some degree.

The last time I dipped my toe into using generative AI in interior design was back in 2023, back when its use was a little more niche, and those wielding it were using a platform known as Midjourney. This was known as 'Diffusion Intelligence', Amir tells us, taking your prompt word by word and translating it into pixels, without comprehension or analysis.

With the advent of multi-modal AI assistants like Gemini, it's become a lot more accessible, affordable, and reliable to use AI image generation. Nano Banana Pro attempts to understand, analyze, and contextualize what you're trying to achieve

an ai scandi style dining room with limewash walls and wabi sabi furniture

(Image credit: Google Gemini / Nano Banana Pro)

During the workshop, we tackled three main tasks. The first: take an image of the room we were sitting in, and apply a new design scheme inspired by your personal style to it. The second: furnish a floor plan schematic with furniture, before applying textures and colors to that floor plan. The third: turn that floor plan into a three-dimensional render.

However, what these were all in aid of was not just showcasing what was possible with Gemini (Amir also showed us some examples of how these renders could be animated into videos, too, which gives them even wider applications), but helping us learn how to prompt effectively.

Because that's where the real art is in AI image generation. The generated output is only as good as what you put into the prompt. So, if it's something you've tried with limited success, or something you want to try for yourself, as an interior designer or just someone designing your own home, here's Amir's advice for how to prompt visualizations for interior design, and some examples of what you can do with this AI tool.

But, just one last note before we get into it. I signed up to Gemini Pro before the workshop to streamline the process. Nano Banana Pro is available to all Gemini users, but if you're on the free version, you'll only get a handful of Nano Banana Pro generations before it automatically switches to standard Nano Banana. For AI Pro users, they get up to 100 Nano Banana Pro generations a day. The image resolution is also much better on the Pro tier, and you can only have three reference images on a prompt in the Free mode, but five or more in Pro.

How to Prompt for AI Interior Design

For those who haven't prompted much in AI before, it can be hard to know how much detail exactly to give. The answer is: the more the better, really, but that doesn't mean you can't get results from a limited prompt output.

However, Amir has a formula to work with that isn't too intensive but should help deliver the right results.

[verb] + [primary subject] + [style/quality] + [details] + [context / environment] + [negative prompt] + [format / aspect ratio]

I'll come back to an example of how this looks, based on one of the workshop tasks shortly, but by keeping to this format, you'll not only make sure you include all the details that you ideally need, and keep it organized.

Inserting Real Life Products Into Your Visualization

If you want to see what a product looks like in your home before you buy, or visualize an interior from a selection of products on your moodboard, you can prompt Nano Banana to do so. Remember when I said that you could add reference images into your home prompt, too? That's an easy way to try this task, but it's not foolproof.

Amir also suggests including a written prompt for each piece in your prompt, too. You can do this, either by describing it yourself or by asking Gemini Pro, in a different conversation, to create a prompt for the reference image that you can then plug into your prompt for the room visualization.

a scandi style room with limewashed wall, windows, and a sofa in the foreground

(Image credit: Google Gemini / Nano Banana Pro)

With that in mind, here's the prompt I used for the image, above:

Create an image of a luxury dining room interior with cornicing, using a modern, wabi-sabi, minimalist design, use the Invisible Collection’s Montil Dining table from the referenced image (using the prompt: A contemporary, sculptural wooden dining table on a stark white background. The tabletop has an organic, asymmetrical pebble shape made of light blonde wood, featuring a distinct, hand-carved, dimpled texture in the center surrounded by a smooth outer border. The table is supported by three mismatched, geometric legs: one wide, flat leg made of dark walnut wood; one wide, flat leg made of matching light blonde wood; and one simple, cylindrical wooden pole. High-end artisanal furniture design, studio lighting), with Moca Dining chairs by Studio Rig from Monologue London (from the other reference image), include a large, textural plain rug, the walls and ceiling are a limewash, textured plaster finish in a slightly warmer, neutral brown, the windows are large sash windows with lightweight linen curtains on a concealed track, and the floor is a wabi sabi textural oak. An oversized, round, minimalist, Japanese fabric pendant light is above the dining table. The table is only set with a sculptural vase with a Japandi-style branch. A large framed canvas is on the wall in a minimalist, wabi sabi design in a muted dark red. A large black olive tree is in a distressed pot in the corner of the room. An oak plinth with a wabi sabi, minimalist sculpture sits next to the wall art. In the foreground, there's the back of a light beige boucle-textured sofa, similar to King Living's 1977 sofa, referenced. Lighting is soft and indirect, with the feel of a high-end residential apartment in London. There should be no people in the image, and no recessed lighting in the room. The image format should be the same as the original image.

Creating a Floorplan in Gemini Pro

a floorplan schematic

(Image credit: Google Gemini / Nano Banana Pro)

Let's be honest, there aren't many people who aren't interior designers who would choose to create a floor plan for their schemes, but it's a nice trick to be able to pull out of the bag, especially if you're trying to convince your other half that your living room layout works.

If you have an existing floorplan schematic, you can start there, or even hand-draw the room, upload it as a reference into Google's AI, and have it turned into one.

Then, we can turn to our formula for prompting and create a new design for the space. If you want to get technical with the scale, that's possible, but here's a quicker prompt to give me a rough design for the small apartment floorplan I created, above.

Create an image using this floorplan as a reference, creating a cad style 2d floorplan without changing the main plan. Add a sunken conversation pit design in the top left area. Add a loft bed in the top right side room, against the right-hand wall vertically, with a desk underneath — this should be layered underneath the bed. In the kitchen, located at the bottom, change the cabinets from a corner L-shape to a U-shape kitchen, connecting new cabinetry to the existing layout. Keep the aspect ratio and format the same as the reference image.

I tried a few different prompts to try to get the loft bed element to work, but this level of layering never quite seemed to compute for this floorplan.

Adding Textures, and Virtual Re-Upholstering

an ai floor plan with rust color carpets, and furniture in bedroom and living room, a teal kitchen, and a chartreuse green living room

(Image credit: Google Gemini / Nano Banana Pro)

The next job was to turn this layout into a floor plan filled with texture, color, and pattern. I had an idea, channeling monochromatic, textural style across the space (think a Dimore Studio vibe).

Again, to do this, I turned to the prompt formula, as well as asking Gemini for some prompts for particular textures for the space, including the carpet for the floors and the velvet for the sofa, using a reference image, too.

Here's how that prompt looked:

Create an image to turn this floorplan into a full 3d textured floorplan with realistic furniture and materiality. Use a rust colored textured carpet for the living room and the bedroom, using the reference carpet image (using prompt A close-up, top-down view of a short-pile carpet texture. The color is a warm, earthy blend of terracotta, burnt orange, and rust. It features a fine, organic, mottled pattern with darker brown stippling and speckles distributed unevenly across the surface. Flat, even lighting, photorealistic material texture, seamless design) but not in the bathroom or kitchen. In the sunken conversation pit, in the top left, use the second texture of a rust colored velvet, matching the tone of the carpet. In the kitchen, uses a seamless teal colored stone, using the reference, a richly veined, bright teal colored marbled surface. The sink should be in the same material, and the range top should be integrated into the teal material. The floor in the kitchen should be ceramic teal tiles. The bathroom should be a bright, chartreuse green floor tile, with a solid chartreuse material bathtub, in a corresponding shade, and a matching basin. In the bedroom, the bedding should be a rust colored silk, matching the color of the carpet, with a rust colored bed frame in the same material as the conversation pit. The desk should be a built-in design, with the surface of a rust-colored tile. Keep the aspect ratio and format the same as the reference image.

a sofa in a patterned pink fabric

(Image credit: Google Gemini / Nano Banana Pro)

It's a similar concept that can be used to re-imagine a piece of furniture in a new way, too. Above is an example of how a Loaf Curvella sofa could be recovered in this Elsworthy fabric design by Harlequin.

Turning Floorplans Into 3D Renders

an ai created room with a rust colored conversation put in an apartment block

(Image credit: Google Gemini / Nano Banana Pro)

The last part of the workshop focused on turning the 2D render into a 3D scheme, and I'll be honest, I ran out of time a little on this one. I could quite get it to realize the walls dividing up the space, and get a hold on the elements it was creating itself. Still, here's a look at the prompt I was workshopping.

Create an image of an eye angle pov of this living room using the floorplan image referenced, enclosed on both sides by walls, and taking note of the position of the patio doors and windows, in the style of high-end editorial interior shots. Perspective render. wide-angle perspective. realistic outside view with a balcony, in a rust colored tile. Beautiful sunshine, realistic sky, and realistic views. Pendant light above the ottoman in the center of the space, using the reference (A modern, sculptural pendant light hanging against a solid white background. It features a stack of three glossy, rounded, ceramic-like donut shapes threaded onto a thin black cord. The top and bottom shapes are a golden mustard yellow, while the middle shape is a terracotta burnt orange. The light hangs from a matching terracotta ceiling canopy, and a frosted white spherical light bulb sits at the very bottom of the stack. High gloss finish, studio lighting.) The textured rust-colored carpet can be created with this referenced texture, and the velvet for the sunken conversation pit, using the reference image for the velvet fabric. The sunken conversation pit should be a U-shape on the left-hand side, lowered slightly from the main floor level, with steps on the right-hand side. There should be a small ridge, also carpeted, opposite the U-shaped seating. In the center, there is a large square ottoman. There should be circular, flat throw cushions on the sofa, spaced evenly. The walls should be a textural, limewash finish in a color match to the carpet. The windows should be color-matched steel in the same color as the walls. Lighting is soft and indirect, with the feel of a high-end residential apartment in London. There should be no people in the image, and no recessed lighting in the room. The image format should be the same as the original image.

What Generative AI Doesn't Do So Well

It's fair to say that generative imagery still isn't the perfect tool, and there's nothing like having a room full of people all using it at the same time to give you evidence of that — you'll get a few glitches every now and again, and sometimes a part of the prompt is ignored, perhaps.

But the main limitation in using something like Google's AI image creation is self-imposed, however. AI doesn't do repetition, so any time you want to slightly tweak a result, you're likely to either end up with a different image being created from an almost identical prompt or you'll feed your existing image back into the AI prompt, but you'll then be looking at a reduction in quality.

But, while it can be a little frustrating to get to grips with, there's no doubt that it's an incredibly powerful tool for visualizations, and one that's going to be the new industry standard. Seeing some of the work interior designers in the session (particularly those who had experience using it before) were generating, it's hard not to be impressed.


Whether or not I'd continue to use Google Gemini for visualizations like this as a non-interior designer, it's hard to say. Since having the Gemini Pro account set up, I used it to insert a rug and two wall lights I was considering into my living room scheme, to see what they'd actually look like. It's hard to say it's not helpful.

I guess I'm torn. Part of the joy I take in interiors is something arriving and working like the vision that, until then, I only had in my mind. At the same time, I also hate having to return things that don't look how I expected when they turn up.

Hugh is Livingetc.com’s editor. With 8 years in the interiors industry under his belt, he has the nose for what people want to know about re-decorating their homes. He prides himself as an expert trend forecaster, visiting design fairs, showrooms and keeping an eye out for emerging designers to hone his eye. He joined Livingetc back in 2022 as a content editor, as a long-time reader of the print magazine, before becoming its online editor. Hugh has previously spent time as an editor for a kitchen and bathroom magazine, and has written for “hands-on” home brands such as Homebuilding & Renovating and Grand Designs magazine, so his knowledge of what it takes to create a home goes beyond the surface, too. Though not a trained interior designer, Hugh has cut his design teeth by managing several major interior design projects to date, each for private clients. He's also a keen DIYer — he's done everything from laying his own patio and building an integrated cooker hood from scratch, to undertaking plenty of creative IKEA hacks to help achieve the luxurious look he loves in design, when his budget doesn't always stretch that far.