AI interior visualisation has been the topic that will not go away for two years. The tools have got better. They have not got as much better as the marketing suggests. After a month of using the major options on real client projects, the honest picture is mixed.

What the tools actually do

There are three job-to-be-done categories. The first is style transfer — take a photo of a room and render it in a different aesthetic. The second is replacement — swap out furniture or finishes while keeping the rest of the room consistent. The third is generation — produce a room from a sketch or prompt with no source photograph.

Style transfer is now reliable. Replacement is patchy. Generation is impressive in demos and rarely useful for client work.

What they fail at

The single biggest failure mode is consistency. Run the same prompt twice and you get different results. For client presentations where you need three angle variations of the same scheme, this is a structural problem. The tools are not stable enough to produce a consistent room across multiple views.

Lighting is the second failure mode. The tools can render light convincingly in isolation, but they do not understand light physics the way a real designer does. Wall colour will look wrong because the light bouncing off the floor is wrong. Furniture shadows will not match the window placement. You catch this in five minutes, and a client may not notice on a single image but will notice on a walkthrough.

The third is materials. Stone textures look photo-real but the veining patterns repeat. Timber looks right but the grain direction is sometimes physically impossible. Fabric weaves are convincing until you zoom in.

What they are useful for

Mood boards and client conversation starters. This is genuinely valuable. I can put together a scheme conversation with a client in twenty minutes that would have taken me half a day three years ago. The decisions get made faster because the client can see options rather than imagine them.

Cost-saving on early-stage concept work. If a client wants three different directions explored, AI tools let me show them three rough directions in an afternoon. The detailed development work is still manual, but the convergence on direction happens faster.

Where I still pay for the human work

Final presentation renders, photo-realistic visualisations for builder briefings, technical drawings — all manual. The AI tools cannot do these to a standard that survives close inspection, and they generate inconsistency that introduces risk into the build.

The Team400 conversation

For studios thinking about how to integrate these tools into their workflow without breaking their delivery quality, I have found firms like Team400 useful for the broader workflow conversation. The interesting work is not picking the right tool — it is figuring out where in the design process the tool earns its place and where it does not. That is a design ops conversation, not a tool selection conversation.

What I would tell a designer starting now

Pick one tool. Use it for mood boards and early concept conversations only. Do not use it for final presentation. Do not promise clients it does things it cannot do. Reassess in six months. The tools will be better. The principles will not change much.