How to Create AI Architectural Renders: From Sketch or Massing Model to Client-Ready Image

A browser and a rough idea. Vision works from any of these starting points:

• Words only: a written brief such as "six-storey mass timber residential block, leafy street, golden hour"
• A screenshot of your model: SketchUp, Revit, Rhino or any massing tool. A flat clay screenshot is enough
• A sketch or photo: hand sketches, site photos, or an earlier render you want to push further

You do not need materials, lighting rigs or entourage set up in the model. Those arrive through the brief. What the reference image contributes is composition and geometry; what you write contributes everything else.

1. Open Vision and choose Exterior render or Interior render, or pick a matching starting point from the gallery, which seeds the tool and brief for you.
2. Attach your screenshot or sketch with Image refs.
3. Set the preservation level: strict keeps your geometry almost untouched, balanced allows the image to breathe, free lets the render reinterpret more freely. For client-facing work, strict or balanced keeps the design honest.
4. Generate. The first result is a direction, not a verdict.

If the geometry is right but the mood is wrong, that is a brief problem, not a model problem. Adjust words, not the model.

Architects already know the vocabulary; the skill is ordering it. A reliable structure is subject, context, light, material, life:

• Subject: what the building is and its massing ("courtyard block of twelve flats")
• Context: where it sits ("conservation area street of London stock brick terraces")
• Light: time and weather ("overcast morning, soft shadows")
• Material: the palette that carries the design ("handmade red brick piers, charred timber upper storey")
• Life: entourage ("people on the pavement, a cyclist passing")

Vision's Add detail chips cover each of these families, so you can assemble a professional brief with taps instead of typing. Add exclusions for anything you do not want: no cars, no scaffolding, no lens flare. Keep it specific and short; thirty well-chosen words outperform two hundred vague ones.

This is where most time is saved. Every result in Vision carries its own next steps:

• Edit with instruction: "swap the facade to red brick", "change to dusk", "add street trees". The rest of the image stays put
• Region editing in the workspace: brush over the part you want changed and describe the change; everything outside the brush stays pixel-identical
• Remove object: parked cars, signage, scaffolding, gone cleanly
• Reroll: same brief, fresh variation, for when the composition is right but the detail is not
• Use as source: promote any result to be the reference for the next iteration

Work on one image until it is right. A render you recognise your own design in is worth more than twenty loosely related ones.

Client presentations fall apart when every image looks like a different project. Consistency comes from reusing three things across the set:

1. The same core brief: copy it and change only what varies (the view, the time of day)
2. The same style preset: Apply style carries a look across renders from different sources
3. The same seed and engine: the seed is recorded with every result, so a look you like can be reproduced deliberately

The result is a board of images that read as one project, one atmosphere, one argument.

For boards and reports, resolution and framing matter as much as the image itself:

• Aspect ratios from 1:1 to 21:9 cover everything from an Instagram square to a full-bleed presentation spread
• Upscale 2x or 4x takes the chosen image to print quality
• Variations grid produces four directions at once when you want options to discuss
• Flythrough video turns a still or a brief into a short cinematic clip for consultations, bids and social channels

When the image is final, Add to Report files it into the same evidence pack as the rest of your site analysis, so the visual and the argument stay together.

Vision is not only photoreal images. The same workspace produces the diagrams a planning conversation needs:

• Masterplan diagrams: illustrative site plans with landscape and movement structure
• Schematic floor plans: early test-fit thinking in plan form
• Site analysis diagrams: labelled sun, wind, views and access
• Figure-ground and sun path studies: computed from real site data, accurate enough for pre-application documents
• Custom map styling: illustrated maps in your presentation's visual language

Because the computed tools draw on the same site data as the rest of Atlasly, the diagram you show is the site you analysed, not an artist's impression of a generic one.