Feature
Last updated
Describe a room and see it
The one mode that needs no photograph. Type what you want and get a photorealistic interior built from the description.
Generate a room from textNo photo
the only mode that needs no input image
26 room types
from entryway to pool house
36 styles
as a starting point for the description
Every other mode starts from a room that exists. This one does not, and that is the whole point of it. There is no source photo, no architecture to preserve, and no before to compare against. You describe a space and the model builds one, which makes it the right tool for concepting, for mood boards, and for the early stage where nobody has committed to a floor plan yet. It is also the wrong tool for redesigning your living room, and knowing which side of that line you are on saves a lot of wasted credits.
What this mode does
- Input
- A room type, style, and written description. No source image is required.
- What changes
- Creates a new conceptual interior from the description.
- What stays
- Nothing from a real room, because no photographed room is supplied.
- Use another mode when
- Use Restyle for a room that exists, or Sketch to Render when a drawn composition exists.
When generating from text is the right call
The honest answer is: when there is no room to photograph. That covers more situations than it sounds like. A buyer deciding between two properties they have only seen listing photos of. An architect concepting a space that does not exist yet. A designer building a pitch deck before the first site visit. A homeowner planning an extension. In all of these the alternative is a mood board assembled from other people's rooms, and a generated set is more coherent than a collected one.
It is also the right call when the room type itself is the question. Someone planning a home gym or a wine room usually has no reference for what a good one looks like, and browsing generated examples is a faster way to develop taste than searching. The vocabulary you acquire doing this is directly reusable: the words that produced a good generated room are the words that will produce a good restyle later.
What it is not for is redesigning a room you own. It is worth being blunt about this because it is the most common misuse. A generated bedroom that looks like yours is not yours: the window is in a different place, the proportions are subtly wrong, and you cannot show it to a builder or a buyer as a proposal for the actual space. If you have a photograph, use a mode that takes one.
Writing a description that produces something good
The difference between a weak and a strong description is specificity about physical things. Warm and inviting is an adjective pair that could describe a thousand rooms and constrains almost nothing. Warm white oak floor, cream boucle sofa, unlacquered brass wall lights, late afternoon light from the left names materials, colours, objects, and a lighting condition, and the model has something to build with.
Materials carry the most weight, followed by light, then colour, then arrangement. Naming a specific wood, metal, or fabric does more for the result than any number of mood words. Specifying the time of day and the direction light enters from does more than most people expect, because it sets the entire tone of the image. Arrangement carries the least weight, and instructions about what sits where are the ones most often quietly ignored.
The other rule is one change at a time. A description asking for six specific things usually delivers three of them well, and if you rewrite the whole thing between attempts you cannot tell which part was responsible. Change one clause, run it again, and you learn something. The prompt limit is 1000 characters, which is far more than a good description needs.
The limits of generating from nothing
A generated room is plausible rather than buildable. Proportions look right without being measured, so a room generated as a small bedroom may not correspond to any real small bedroom, and the furniture in it is not drawn from a catalogue you can order from. Treating the output as a concept rather than a plan is the correct posture, and it is why the mode belongs at the start of a project rather than the end.
The same content rules apply here as everywhere else. No people or animals appear in output. No logos, watermarks, brand names, or text overlays are generated, which means you cannot ask for a specific manufacturer's furniture by name. What you can do is describe the piece: a low-slung leather lounge chair on a walnut base gets you close to what naming a brand would, without putting someone else's trademark in an image you may use commercially.
Finally, there is no before. Every other mode gives you a comparison, and the comparison is most of how you judge whether a result is good. Here there is only the image and your own reaction to it, which makes it worth generating three or four and choosing, rather than accepting the first plausible room the model returns.
Turning a generated room into a real one
Since the output is a concept, the value is in what you carry out of it, and two things travel well. The first is the vocabulary. The words that produced a room you liked are a brief, and a far more precise one than most people manage to assemble on their own. Handing a designer, a joiner, or a furniture search the phrase smoked oak floor, deep green cabinetry, brass hardware, warm low light gets you further than a folder of images with nothing said about them.
The second is the image, as an instruction to a person rather than a specification. A builder or a fitter can read intent out of it: the material, the tone, the density of the room, roughly where things sit. What they cannot read out of it is a dimension, a product, or a construction detail, because none of those were ever in it. Presented as a direction it is genuinely useful, and presented as a plan it will waste somebody's afternoon.
Once a real room exists, change modes. The same description that produced a good generated room can be given to a mode that starts from a photograph, and the result is the same direction applied to the space you actually have, with your windows in the right places. That handoff is the point of this mode: it belongs at the front of a project, and it hands over as soon as there is something to photograph.
Why generate a room from text works
Start before the space exists
Useful when there is no room to photograph: a property you have not bought, an extension not yet built, a concept for a pitch.
Build a mood board that agrees with itself
Generating several rooms from one description gives a coherent set, where images collected from different sources rarely cohere.
No architecture to fight
With nothing to preserve, the model has full latitude. Ceiling heights, window placement, and proportions are all yours to specify.
Test a description cheaply
Working out how to describe what you want is easier when you can see each attempt. The vocabulary you build here transfers to the photo-based modes.
Room types that are hard to photograph
A home gym, a wine room, a pool house: spaces people plan long before they have one to shoot.
Not a substitute for restyling
If you have a room and want to see it changed, use Restyle. A generated room is a different room, however good it looks.
How it works
- 1
Choose a room type and style
Pick from 26 room types and 36 styles as the base of the description.
- 2
Describe the specifics
Materials, colours, light, and the pieces you want in it. Specific materials work far better than adjectives.
- 3
Generate and iterate
Get a photorealistic interior in about 30 seconds. Adjust one part of the description and run it again.
Related tools
Go deeper
Frequently asked questions
Ready to try it on a real room?
Generate a room from text