Haven’t you heard? The age of AI is upon us, though the news might have been easy to miss; it’s not like there are a thousand bearded “disruptors” and hypemen tweeting their way onto every conference stage in the Bay Area to tell us all breathlessly, over and over again, how we’re living through the fourth industrial revolution. Definitely not.
Allow me, then, to introduce you to the newest interior designer on the block: ChatGPT. The Large Language Model (LLM) is an omnitool of a chatbot, one which can do anything from offer you emotional support to help you cheat on your degree. According to its creators at OpenAI, “ChatGPT helps you get answers, find inspiration and be more productive.” So, I did the natural thing and asked it what colours I should paint my flat.
I need some advice on what colour to paint several rooms in my flat, I wrote. Can I attach some images to show you the space(s)?
It is at this point that I encounter my first issue. No, ChatGPT tells me. It can’t see photos, and I can’t upload them, even if I make an account, apparently. Nor does it have access to the internet, if I upload my images somewhere else like Imgur: “I can’t browse or view anything online, even if you upload images somewhere. But I can still help you based on your descriptions. If you tell me a bit about the space—like wall sizes, colors you’re already using, and the atmosphere you're aiming for—I can suggest colors that would work well. Feel free to share what you're thinking!” (In true AI fashion, GPT makes ample use of the em-dash.)
I am surprised by this, but it’s not the end of the world. I decide to give it a go. I’m a writer, after all.
Here’s the description of the first room, the spare room in which my flatmate lived until last week. It’s a north-facing room in Pimlico, in London, on the seventh storey of a block of flats. It has a large ribbon-like window alongside the entire north wall, which starts at the top of the wall and goes down to about halfway. The room is a square, and has built-in floor to ceiling storage along the wall opposite the windows (the south wall). The door is in the south-east corner. It’s also white. The flat was built in the early 1960s (it’s a development called Churchill Gardens) and the carpet is a mid-grey. (I promise I didn’t choose it).
It’s at this point I realise I am justifying myself to a computer.
This is the second bedroom of three in the flat and will obviously be used as a room for sleeping but also probably for working from home. It currently has no furniture in it except for a grey metal double bed frame from IKEA. The frame has a headboard which is quite simple, also in grey metal.
ChatGPT’s cursor blinks a couple of times, and then the response materialises: “Thanks for that detailed description! Sounds like a really interesting space with a lot of potential.”
Not sure about that. “A lot of potential” – am I being patronised by GPT? It’s definitely not an interesting space; the most interesting thing about it is that you can hear every decibel of the scooters the local kids like to rev up and down the street at 2am, even seven storeys up.
“Since it’s a north-facing room,” the AI continues, “natural light will likely be a bit cooler and softer, which can make it feel a little colder or less bright, especially in the winter months. I’m thinking a warmer tone could balance out the cooler light and create a cozy, inviting atmosphere.”
This is smart, if basic-level interior design. The light in here will be colder, and the normal advice for the room would be to avoid too stark a white so as not to stray into the clinical.
GPT follows with several ideas, namely “Warm Beige / Taupe” – which will keep the room “calm, cozy and not too stark”, and apparently works for a minimalistic or modern vibe – or “Soft Greys with Warm Undertones”, which GPT says will avoid a clash with a grey carpet (I’m unsure about that). Sage green or soft olive will apparently likewise pair well with a grey (true), and green is apparently helpful with creating focus – which might be helpful for working from home, it says. A dusty rose or a peach is perhaps the most intriguing of the suggestions; also good with grey, and warm for a north-facing room.
Finally, GPT suggests, if I want “a little more drama” or a “Bold Contrast” (the AI slightly shoots from the hip when it comes to capitalisation), I might try deep charcoal, navy or even a “moody forest green” as an accent wall. “You could then layer in accents of gold or brass through furniture or decor to add warmth and elegance.”
I know for a fact that I’m not going to paint the room forest green, realistically, but I’m interested in talking to GPT more about colour. I quite like the idea of forest green, I write. The problem is, I want the room to feel interesting and not soulless, but I can't guarantee who will live there in the future so it can't be *too* out there.
The solution comes back quickly: GPT suggests an accent wall again. If I don’t want too dramatic a colour, it tells me, I could just paint one wall, “ideally the one opposite the window (the south wall). This will draw the eye, create a sense of depth, and still leave the room feeling open. The other three walls can remain a lighter neutral (maybe soft grey, warm beige, or off-white) to keep the room feeling bright and spacious.”
At this point, I remember I have a GPT account associated with my work email. I don’t trust GPT’s assertion that I can’t upload an image even if I log in, so I find the password for my old account and log in. Can I upload a picture of my sitting room for you to give me advice on what colour I should paint it? I ask this new iteration of the chatbot.
The reply is instant. “Absolutely! Please go ahead and upload a picture of your sitting room. Once I can see the space, lighting, furniture, and overall vibe, I’ll give you tailored paint color suggestions that suit the room and your style.”
Brilliant. I have no idea what the first instance of GPT was thinking, but it was totally wrong about uploading images. I scamper downstairs to take a picture or two, stopping to move some washing and put the ironing board away (reader, you are not truly houseproud until you have tidied your flat for an AI).
This time, I don’t tell it anything about the room in words – not that it’s south-facing, what the flooring is made of, nothing. I just upload three photos. GPT immediately responds. “Thanks for sharing the photos — your sitting room has a lovely warm mid-century-modern feel, with great natural light, lots of wood tones, and a nice balance of textures.”
.jpg)
.jpg)
This time, GPT recommends specific paints from Dulux and Farrow & Ball – oddly enough, it picks one named paint from each brand, each time it suggests a colour combination. For warm neutrals, for example, it suggested “Skimming Stone” by Farrow & Ball or “Egyptian Cotton” by Dulux. These, apparently, work well with “wood tones, rattan, black metal accents and [my] leafy plants”. But I should avoid anything too yellow or “it might clash with the softer wood tones.” I actually disagree with this. I like a serious yellow room, which can look beautiful in the mornings and evenings, and of course in the autumn. The first flatshare I lived in in my 20s had an inspired and unusual deep yellow sitting room, in which we spent many peaceful hours and which was well furnished with wooden pieces – albeit perhaps a tone or two darker than my sideboard.
GPT’s next suggestions for this room are greens, which “add depth and connection to the plants and wood”. I should think about “Lichen” by Farrow & Ball, apparently, or “Tranquil Dawn” by Dulux. I agree that one of these would go with the mid-century sideboard and the plants, at least in theory. Out of curiosity, I ask if GPT gets a commission if someone buys a paint it recommends. It doesn’t, it says, and “can also suggest options from specific brands you prefer, or give you a palette that works with budget-friendly paints (like B&Q, Wilko, or Valspar) or premium ranges (like Little Greene, Farrow & Ball, etc.). Just let me know your preference!”
It’s a kind offer but instead, I ask how it would wallpaper the room if I so desired (I am imagining some sort of ChatGP&J Baker). This time, I think the bot misses the mark badly. Its first advice? You guessed it – feature wall! “You don’t have to paper the whole room — often one feature wall is all you need to transform a space.” It suggests that the wall behind the dining table would be the best candidate. I strongly disagree that that wall is the best to wallpaper in the room, but more importantly, I don’t think a single wall of wallpaper is a good idea. It might be fine in a very dressed-up room with many different textures and patterns – one which had been very consciously designed – but in my flat, it would just look like the papering was unfinished.
As for style and pattern ideas, they’re a little more comprehensible. As it is wont to do, GPT gives me three options: Botanical, Mid-Century Geometric or Textural wallpaper. It also recommends what to look for in each: “William Morris-inspired prints or modern leafy patterns” if I want to go botanical, for example, or prints from Sandberg, MissPrint or Ferm Living for mid-century styles.
At this point, GPT offers to show me a mocked-up version of my room with different wallpapers. I ask it for a geometric and a foliage pattern, and it gets to grinding out an image. I say a silent prayer for the acreage of rainforest being consumed to power my whimsical request as I wait for the image to generate. When it arrives, I realise it has misunderstood my ask – rather than give me one mock-up with a geometric pattern, and one with a botanical print, it has given me a geometric botanical print (albeit one which appears not to follow any legible logic).
Nonetheleess, I’m impressed by the render. It looks convincing, almost like the rich green walls of a Victorian picture gallery, though I am newly convinced that I don’t want to live in a jungle.
At the other end of the room, I’ve long wondered whether I should fill in a gap above the TV which feels a little like a white void in the room (though the blank wall was very useful for watching films on a projector when we first moved into the flat, before we owned a TV). In writing, GPT offers me some solutions, including mounting a single open shelf or creating a “gallery wall” of several pictures. I ask it to mock an option up for me, and it does.
It looks nice enough. This is the best decoration suggestion so far, disregarding the slightly odd positioning of two frames behind the TV, and the lack of symmetry (I think the lampshade to the right of the TV has thrown the AI off a little). It makes me want to know: where is GPT getting all these ideas? “While I don’t have access to real-time data by default,” it explains, “I’ve been updated with design trends up until 2024. That includes things like the rise in peel-and-stick wallpaper, sage green as a dominant tone, and the return of mid-century modern and natural textures.” Aha – so that’s where the consistent suggestions of sage and forest green keep coming from.
And why does it keep suggesting me accent walls? “I don’t suggest accent walls by default,” says ChatGPT, “but in your case, I keep circling back to them because of a few specific things I noticed in your room.” These, apparently, are “focal points” – like one wall at the end of the room which has a print of an upside-down map of London on it – which lend themselves to an accent wall.
All fascinating, and some of it useful. I’m not sure that interior designers need to be too worried about being replaced any time soon; by definition, anything that ChatGPT can tell you can also be found somewhere on the internet or among its training data, meaning much of its advice is already public (including a lot published on House & Garden’s own website). And some enterprising designers are using AI themselves already. That said, if you find yourself yearning for something bold and needing validation – perhaps an accent wall in forest green? – then you’re sure to be in good hands.
