NewCrit Space is open to everyone. Invite people to your project.Open a project

Architects Are Firmly in Control of AI: 98% of Renders Start From Their Own Work

Shaun McCallum

Shaun McCallum

November 17, 2026

What our latest render data says about how architects and interior designers work with AI.


Architects and interior designers are firmly in control of AI. In the last three months, 98% of the renders made in Fenestra started from the designer's own model, drawing, sketch or photo. Fewer than 2 in 100 started from a text prompt alone.

Black timber house at dusk with warm light behind the glazing, rendered from a view of a Rhino model

I worked as an architectural designer before I built software, and the question I still get asked most about AI is whether it is going to design the building for us. I understand the worry. Most of what gets shared online is a sentence typed into a box and a finished tower coming out the other side.

It's not what we see in Fenestra. People upload a view from their model, a plan or a photo of a site, and ask for materials and light. So we looked at the numbers to find out whether that was a few careful people or nearly everyone. It is nearly everyone.

How many AI renders start from the designer's own work?

98% of renders in Fenestra begin from an image the designer supplied. In practice that is a screenshot of a model, an exported drawing, a scan of a sketch or a photo of a room.

98%of renders start from the designer's own image
Share of renders in Fenestra over the last three months.

The share has gone up in each of those three months. People who design buildings don't type a sentence and wait for a building.

What do architects start an AI render from?

Mostly a 3D model or a CAD drawing. We ask every new member what they will start from.

48%A 3D model
What new Fenestra members say they start from.

Four in five arrive with their own model or drawing. About one in ten keep the 3D model inside their Fenestra project. The rest bring a view of it as an image, so the modelling stays in Rhino, SketchUp, Revit or Archicad, where the decisions are made.

Here is what each starting point looks like.

A 3D model

Untextured view of a house modelled in Rhino
Render of the same view as a black timber house in overcast daylight
A view from a Rhino model, and the render made from it.

The massing, the openings and the camera angle all come from the model. More in rendering a Rhino model with AI.

A 2D CAD drawing

Flat 2D floor plan used as the input
Top down render of the same plan with timber flooring, furniture and long shadows
A floor plan, and the same plan rendered from above. The walls and rooms stay where they were drawn.

The render follows the layout as drawn. More in 2D plans to interiors.

A photo

Kitchen with pale cabinets, a white stone island, an oak slatted ceiling and a terrazzo floor
The same kitchen with a green marble island, a coffee machine on the back counter and two people
A site photo, and the same kitchen with the island changed.

The designer picks what changes and what stays. More in AI kitchen design from a photo.

A sketch

Hand sketch of a building used as the starting image for an AI render
AI render that keeps the form and proportions of the sketch
A hand sketch, and the render that keeps its form and proportions.

The form and proportions come from the sketch. More in sketch to render.

Which AI models do architects use most?

One model does close to half the work. Nano Banana 2 makes 46% of all renders and has been the most-used model every month since March. Nano Banana Pro makes about a fifth.

  • Nano Banana 246%
  • Nano Banana Pro18%
  • Muse Image Edit10%
  • GPT Image 2.5 Sunburst and Flare5%
  • Nano Banana 2.13%
  • Other18%
Share of renders in Fenestra over the last three months.

New models arrive most months and get tried. None has come close to the leader.

We read that as a preference for a model you can predict. When a client is waiting on a view, people pick the model whose behaviour they already know. The newer ones are worth a look for specific jobs: Nano Banana 2.1 for composing from references, Flux 3 Klein for quick box edits.

Do architects render through AI agents like ChatGPT and Claude?

Yes, and the same rule holds. Fenestra connects to ChatGPT and Claude, so a designer can brief the assistant in a chat, attach a view and get the render back in the conversation.

The assistant does the asking and Fenestra does the rendering, but the image still comes from the designer. If there is no image in the chat, Fenestra shows an upload card and waits for one. The assistant never invents the starting point. Setup is in the ChatGPT and Claude guide.

Does this mean AI will replace architects?

We don't think so. In 98 of 100 renders the design exists before the AI sees it. The massing, the plan, the openings and the proportions are already drawn. The AI is asked for materials, light and atmosphere on top of a decision somebody qualified has made.

We think they work this way because they want to stay the author, and because a client pays for their judgement. We wrote more on the wider question in Will AI replace architects or enhance them?

Where does this data come from?

Renders made in Fenestra from September to mid-November 2026, and what members who joined in that period told us they start from. We update this page as the numbers change.

Start from your own model or drawing

Open Fenestra

Frequently asked questions

They start from their own work. In the last three months, 98% of renders made in Fenestra began from an image the designer supplied, such as a view of a 3D model, a CAD drawing, a sketch or a photo. Under 2% began from a text prompt alone.

Rarely on their own. In Fenestra's latest data, fewer than 2 in 100 renders started from a prompt with no image. The prompt describes materials, light and mood for a design that already exists.

Asked when they join Fenestra, 48% said a 3D model, 31% said 2D CAD drawings, 12% a photo and 3% a sketch.

In Fenestra, Nano Banana 2 makes close to half of all renders and has been the most-used model every month since March 2026. Nano Banana Pro makes about a fifth and Muse Image Edit about a tenth.

The data points the other way. Architects and interior designers bring a finished design decision to the AI in 98% of renders and use it to visualise that decision. The design is theirs before the model sees it.