Ask architects what worries them about AI and the answer is rarely their jobs. It's the buildings. It's the floor-plan generators and one-click renderers that flatten design into interchangeable, developer-grade boxes, waved through for the sake of speed.

Bryant Callahan has heard that fear plenty. He runs the tech studio at LMN Architects, and his response runs the other way. "We don't want to compress so we can do more projects. We want to improve the quality of work on the projects we have. This lets us actually design."

That tracks with how the Seattle-based LMN operates. They were early on 3D printing, CNC fabrication, and computational design, usually before the rest of the industry moved. "Architecture and engineering tends to lag," Bryant says. "It's so governed by construction, and construction is risk-averse." LMN's instinct runs the other way: try the thing while it's still a little risky, then work out what it's actually for.

LMN can afford that instinct because of what it's achieved: nearly 300 design awards, the 2016 AIA Architecture Firm Award, and a portfolio that runs from the Washington State Convention Center expansion to the Seattle Aquarium's Ocean Pavilion. Architecture, urban design, landscape, interiors, and public art all sit in one studio and turn up on every project. Clients come to LMN when a building has to carry a lot of cultural weight at once, which as it turned out, put a lot of pressure on rendering.

Clients see something as it is

Civic projects generate an enormous amount of visualization. Schematic design runs for months, and at every milestone a team has to turn a still-changing idea into something a city council or a museum board can react to.

For decades, the process worked one way. You designed in 3D, modeled in Enscape or Revit, then late and under deadline, two specialists in 3ds Max produced a polished image for the presentation. It worked, but it also created three problems:

Rendering came too late to act on
Lean teams couldn't afford it
Early polish froze decisions before they were made.

"Clients can immediately start asking, why are the lights like this? Why is the ground like this?" Bryant says. "Designers understand everything is malleable. Clients see something as it looks, in that moment."

A finished-looking render at the schematic stage pulls the conversation toward finishes and polish, when it should still be about feeling, scale, and intent. LMN had been chipping at this for years, cycling through all the node-based tools (ComfyUI, etc) and Midjourney back when it was a Discord channel. None stuck. Local models were unstable under deadline, and most generative tools were black boxes: type a prompt, get an image, which is the opposite of how a design team thinks.

Why FLORA held up

Two things made the difference, and both are less glamorous than they sound.

First was reliability. "We're at a deadline," Bryant says. "You want something that opens up and just runs." It's the boring prerequisite for any tool entering a professional workflow, and where most generative tools fall down.

Second was fit. FLORA adapted to how the studio already worked, instead of the reverse. Jeff, a designer in the studio, became one of the first power users by sketching instead of typing prompts, using a hand drawing as the fastest way to tell the tool what he meant. When the team flagged that copy-paste was clunky, it got fixed. When Jeff's sketch-over method took hold, FLORA leaned into it.

"Picking software often feels like picking the least-worst option," Bryant says. "It's nice to feel like our input had an impact, and that there's recognition of the unique way we work."

Feeling the magic of FLORA

Jeff's process is a loop. He exports a clay model from Enscape, sketches over it (by hand, in Photoshop, sometimes with a finger on his phone) to suggest material, light, or a specific move, then feeds it into FLORA and prompts through a few iterations. Sometimes it ends there, with an internal render that gets the team aligned. Other times it runs back through Photoshop and FLORA again before going to the client. "It's not a linear workflow," Jeff says. "There's a start and an end, but a lot of looping back in between."

Bryant frames it as "multimodal by design": model, export, sketch, render, print, mark up, re-model. "You gain design insight by changing from one form to another and back again."

Here's what that looks like in practice. As LMN plans its own new office renovation, the team had a rough working model of a new layout, the kind of massing and furniture study that makes sense to the people who've been staring at the project for weeks and means almost nothing to anyone else. Team members ran a basic render capture through FLORA, and in about half a minute, that generic model became a rendering anyone could read at a glance.

What has made this use case interesting is that LMN was both designer and client at once. The team got to sit on both sides of the table in a way that almost never happens on a real project.

As designers, they were pushing the layout forward, testing instincts, deciding what was worth pursuing. As the client, they were the ones seeing the render cold, reacting the way any occupant would, with the same skepticism and gut-level sense of whether it actually worked for them that clients bring.

Watching those two reactions happen in the same room, sometimes with the same person, says a lot about why the tool matters. It's not just about the designer moving faster, it's about closing the gap between what a designer sees in a rough model and what everyone else needs to see to respond to it.

That render went to firm leadership as a rough sketch, not a finished proposal. It opened a real conversation, and that conversation told the team where to take the layout. The project team used that back-and-forth to make real decisions: what to commit real time to, what to drop, what needed an actual model. Once the idea held up, it got shared out to the rest of the office. Used this way, FLORA isn't about replacing the design process. It's about giving that process a way to talk to people who aren't inside it yet, before real time gets spent.

What actually changed downstream

The effects from FLORA are quiet, but real.

Schematic design doesn't finish faster, since there's no shortcut through the actual thinking, but it finishes clearer.
Teams move into the next phase with shared intent instead of three different mental pictures of the same room.
That clarity carries forward: fewer surprises late in design development, fewer revision requests once construction starts.

It also flattens an old hierarchy. Rendering used to be a privilege of teams big enough to staff a visualizer. Now the smallest teams can do it themselves.

"A small team of three people can now get renders for themselves," Bryant says. "They have this thing they can go to and say, I want to generate an image, and the result is actually useful."

That access is what Bryant says he'd miss most if FLORA went away. "It's single-handedly one of the most important impacts on this office. How we're visualizing things, and giving people access to it."

Designing workflows around AI

LMN doesn't describe this as adopting AI. They describe it as building workflows around what AI is good at. "There's a lot of fear in this industry," Bryant says. "People worry it'll replace them."

The studio treats FLORA the way it treats its other computational tools, as an extension of the team's own thinking rather than a button you press to skip the work. "We've always asked how we want to use this," Bryant says. "Not how everyone else is using it."