You modeled it. Now a client wants to see it, and the model looks like a model.
The usual answer is a render engine: materials, lights, an environment, output settings, a preview that takes 40 seconds to tell you the sun is in the wrong place. The other answer, increasingly, is an AI tool with a text box that asks you to describe the building you have already built.
Or point Archivision at the SketchUp view you already have on screen and get a photorealistic render back in about 20 seconds.
What you need
The only input is a SketchUp view.
Export it with File > Export > 2D Graphic, or take a screenshot of the modeling window. Upload the PNG or JPG. There is no plugin to install, no extension to keep current when SketchUp updates, no render engine to license, and nothing to configure inside SketchUp itself.
The same upload works from SketchUp Free in the browser, from SketchUp Go, and from SketchUp Pro, because the render is conditioned on the image, not on the .skp file. Whatever your seat costs has no bearing on what comes back.
How finished does the model need to be?
Massing is enough. An untextured white model with correct proportions renders well, because proportions are the part the render has to respect and materials are the part it supplies. A model with materials already applied renders well too, and holds closer to the palette you chose.
What helps most is a considered camera. Set the view the way you would set it for a client: eye height for a street view, a two-point perspective for a corner, enough foreground to give the building a site. The render inherits your framing, so a lazy camera produces a lazy image.
How it works
Three steps, no settings panel.
- Upload the exported view.
- Pick one of 24 architectural styles.
- Render. About 20 seconds.
Why there is no prompt box
Most tools in this category are chat tools wearing an architecture skin. They hand you an empty text field and make the quality of your render a function of how well you write prompts.
Architects should not have to prompt-engineer. You already did the structured design work, in a structured modeling tool, using structured inputs. Asking you to then describe that work in prose so a model can guess its way back to it is a step backward. Pick a style, get a render, change the style if it is wrong. That loop takes seconds, and it is repeatable, which a prompt is not.
Your geometry stays yours
The honest fear about AI rendering is that you upload a careful model and get back a handsome building that is not the one you designed.
The render starts from your export. It is conditioned on the pixels of the view you uploaded, so the massing, roof pitch, floor lines, window positions, and openings you modeled are the ones that come back. Materials, lighting, planting, sky, and surrounding context are what the render adds, because those are the parts your SketchUp view does not have yet.
Where the fidelity risk is
In front of a professional, an AI render fails on coherence long before it fails on resolution.
The failure that matters is a foreground that reads your building faithfully while, somewhere in the background, the model invents an extra wing, a second entry, or a family room where there should be a hedge.
We treat foreground and background coherence as an engineering problem and evaluate against it rather than claiming it is solved, because it is not solved anywhere in this category yet. Alongside that work, region editing lets you select the part of the image that came back wrong and regenerate only that region, so one bad background does not cost you the whole render.
Iterate before you commit
The change that matters is that iteration became affordable.
A traditional render is a 30-hour bet that the client will like the direction. You set up the scene, you queue it, you present it, and if the answer is "What if it were warmer, and lower," you pay the 30 hours again. So nobody explores. You pick one direction, render it well, and defend it.
At 20 seconds a view, you can bring five directions to the first meeting instead of one: the same SketchUp massing in modern, in farmhouse, in Scandinavian, with a heavier stone base, with a deeper overhang. You align on direction in the meeting, in front of the client, and then you spend the 30 hours once, on the direction they actually chose.
Early-phase images are alignment tools, not deliverables, and they are worth most when there are enough of them to choose between.
Once a direction lands, perspective generation will spawn additional views from the render you already have, so the street view and the interior come from the same design rather than from two separate uploads.
SketchUp AI rendering vs. a traditional render engine
Both belong in a practice. They answer different questions.
| Archivision | V-Ray, Enscape, Twinmotion | |
|---|---|---|
| Input | An exported view or screenshot | A fully materialed, lit model |
| Setup | A style choice | Materials, lights, environment, output settings |
| Time to first image | About 20 seconds | Hours of setup, then minutes per frame |
| Hardware | Any browser | A workstation GPU |
| Best at | Concept, options, client alignment | Final, specified, construction-phase visualization |
| Cost | From $19 a month | Seat licenses plus the machine |
A render engine rewards the time you put into it, and for a final image with specified materials that reward is real. It is the wrong tool for the question "Which of these five directions do we take?" asked at 4 p.m. the day before a meeting.
What it costs
Three plans, and no plan is a trial in disguise.
- Starter, $19 a month: 80 credits, 10 projects, eight styles, 2K downloads.
- Pro, $39 a month: 200 credits, 25 projects, all 24 styles, Make It Real hyperrealism, 4K downloads.
- Studio, $79 a month: 500 credits, 60 projects, priority rendering, API access.
One credit is one generation step: a render, a refinement, a region edit, or an additional perspective. A typical project runs six to eight credits. Every plan carries a 14-day money-back guarantee, and you can change plan or cancel at any time. Full detail is on the pricing page.
We charge from the first render on purpose. A tool that fills up with people who will never pay for it gets optimized for the people who will never pay for it, and that is not the tool a practicing architect wants to depend on.
Where to start
Export one view of the project on your screen right now. Not the hero shot, the one you are least sure about. Run it through three styles and see whether any of them tells you something about the design you did not already know.
If your input is a hand drawing rather than a model, the same pipeline handles it: See sketch to render. If you are comparing tools across a practice rather than for yourself, AI rendering software has the wider view.