Atlasly Vision: AI Rendering Built for Architects, Urban Designers and Planners

Vision is the visualisation surface of Atlasly, available at /vision inside the app. It combines four kinds of capability in one place:

• Renders — photoreal exteriors and interiors from a written brief or a reference image
• Edits — change a render or photo with an instruction, remove objects cleanly, or apply a curated architectural style
• Diagrams and maps — masterplan diagrams, schematic floor plans, site analysis diagrams, figure-ground and sun path studies, and custom map styling
• Output — variations grids for comparing directions, print-quality upscaling, and short cinematic flythrough videos

Everything runs in the browser. There is nothing to install, and no render farm to configure.

The Explore tab organises starting points by discipline, so the first click is already close to your workflow.

Architects typically start with dusk or golden-hour exteriors from a massing model, material studies of a courtyard elevation, or biophilic interiors for client moodboards. Attach a screenshot of the model, describe the atmosphere, and iterate on the same image until the materiality reads right.

Urban designers lean on the aerial masterplan, street-section and map styling tools — illustrative plans that communicate structure, landscape and movement without a day in Illustrator.

Planners use the computed diagrams: figure-ground and sun path studies are generated from real site data rather than imagined, so they are accurate enough to sit in a pre-application document. Site analysis diagrams label sun, wind, views and access for committee-ready context.

Vision rewards the way architects already describe buildings. A short, specific brief beats a long vague one, and the composer helps you build it:

• Refinement chips add camera angle, light, season, atmosphere, materials and entourage with one tap — aerial view, golden hour, board-marked concrete, people on the pavement
• Exclusions let you state what to leave out — no cars, no scaffolding, no people
• Framing and batch controls set the aspect ratio (including 21:9 for presentation boards) and how many options to generate at once

Every tool remembers its own settings, so your interior renders stay portrait and your exteriors stay widescreen without re-setting anything.

Vision routes each tool to a chain of leading engines — including Flux 2 Pro, Nano Banana 2, GPT Image, Gemini, Recraft, Imagen, SeedVR and Kling — and falls back automatically if one is unavailable, so a run succeeds rather than erroring.

You can leave the picker on Recommended and never think about it. If you do want control, the picker groups engines into Best quality and Fast & lighter tiers, with a simple $ to $$$ symbol showing how heavy each one is. Expensive does not always mean better for a given task — a fast engine is often the right choice for early iteration, with a quality engine reserved for the image that goes in front of the client.

A render is rarely right first time, so every result in Vision carries its own next steps:

• Use as source feeds the image back into the editor, so you refine the same view instead of starting over
• Reroll runs the same brief with fresh variation when the composition is right but the detail is not
• Upscale takes a chosen result to print quality at 2x or 4x
• Copy description lifts the exact brief for reuse across a set of images

The workspace view goes further: brush over a region to change only that area, hold the geometry with a preservation dial, compare two directions with a slider, and see exactly how each image was made — brief, settings and lineage included.

This is where Vision differs from a standalone image tool. When you work inside a project, Vision knows the site you are analysing — the composer's site chip shows the address and local authority it is working from, and that context shapes the output.

Finished images do not sit in a downloads folder. Add to Report and Add to Pack place them into the same evidence pack as your constraints mapping, flood risk and planning analysis, so the visual and the argument travel together. A figure-ground study, a massing render and a constraints map can all come from one workspace, about one site.

Open Vision from the Atlasly app, pick a starting point from the gallery that is close to what you need, adjust the brief, and generate. The whole loop — brief, generate, refine, upscale, file into a report — takes minutes rather than days.