The slowest part of rendering was never the render. It was the two days of assigning materials, placing lights and waiting for a machine, so that a client could say "actually, can we see it in brick?"
A viewport screenshot skips that. You take the view you already have on screen — SketchUp, Revit, Rhino, Blender, anything — and an AI model turns it into a photograph of the same building. It takes under a minute and it costs 4 credits ($0.35). Here is how to get a usable one rather than a pretty one.
Set the view before you export
Everything that goes wrong downstream starts here.
- Use a real camera height. Set the eye height to about 1.6 m and turn off perspective exaggeration. A view from 40 m up reads as a model, and the render will look like a model too.
- Two-point, not three-point. Keep vertical lines vertical. Architectural photography does; a drifted horizon is the fastest way to make a render feel fake.
- Give the building air. Leave space around the silhouette. Cropping tight forces the AI to invent whatever it thinks continues past the frame edge.
- Hide the clutter. Section planes, axes, dimension text and guide lines all end up interpreted as architecture. Turn them off.
- Export big. 2000 px on the long edge or more. A small screenshot gives the model less to read, and detail it cannot see is detail it will invent.
Say what the materials are
An untextured screenshot is a set of grey shapes. The AI will pick finishes if you do not, and it will pick confidently. Name them: "off-white render walls, dark-stained vertical timber screen, exposed concrete plinth, standing-seam metal roof, charcoal aluminium window frames". Naming every visible material is the single highest-value habit in the whole workflow, because it is the difference between your building and a building.
The same logic applies to context. If you say nothing about the setting, you get a default: a lawn, a driveway, a mid-morning sky. Say "narrow suburban street, mature trees on the far side, overcast light" and you get something that looks like where the project actually is.
The two checks before it leaves your desk
Count the windows. Literally. Put the screenshot and the render side by side and count openings on each elevation, then check the roof pitch and the floor line. AI renders drift, and the drift is always in the direction of "a nicer, more generic house".
Do not fix it with another pass. If something is wrong, change the brief and re-run. Chaining edits onto an edited image degrades it — contrast climbs, detail smears, and by the third pass it looks like a phone filter. Two passes is the practical ceiling.
What it costs
A photoreal image is 4 credits ($0.35). Rendering a 2D elevation or plan at 1K is 4 credits ($0.35). A new account gets 14 free credits and no card is required, which is enough to run your own screenshot through and see whether the result is worth the workflow change.
When this is the wrong tool
If the project deserves a properly modelled scene — real context, considered light, a matched set of views, a camera that moves through geometry that actually exists — a desktop renderer is still the right answer. This is for the render that otherwise would not get made: the one due tomorrow, from a model that is not finished, on the laptop you happen to have.
Common questions
Do I need to apply materials in SketchUp first?
No, and in some ways it is better if you do not. An untextured model gives the AI nothing to fight with, so what you type is what you get. What matters far more is the camera and the silhouette.
Will it change my building?
It can. Openings drift, a chimney appears, a storey grows. Compare the output against the screenshot before you send it anywhere, and if it drifted, re-run rather than editing again.
What resolution do I get?
1K or 2K depending on the tool and setting, with a separate upscale step for print resolution.
How much does one render cost?
4 credits ($0.35) for a photoreal image, and a new account gets 14 free credits with no card.
Try it on a screenshot you already have: 3D Model to Render.