Renovato/ ai
articleA-006 · tutorial
published2026 · 05 · 06
cluster · 3D

Render to 3D model,
for architecture.

AI image-to-3D for architecture is a class of generative system that takes a single architectural render and produces a textured 3D asset — geometry plus baked PBR materials, exported as GLB, FBX, or USDZ — without going back to the original 3D model. The output is suitable for Unreal, Unity, Blender, three.js, and Apple ARKit; it is not a metric-grade BIM model.

What you actually get out

The output is a single self-contained file: GLB (binary glTF) by default, with FBX and USDZ available as alternate exports. Inside the file: a polygonal mesh (typically 50k-200k triangles for a building-scale output), with PBR material maps baked into the textures — albedo, roughness, metallic, and normal.

Drop the GLB into Three.js, Blender, Unreal, or Unity and the materials behave correctly without manual setup. Drop the USDZ into iMessage and an iPhone user can scale the building life-size in AR. Drop the FBX into a legacy 3D pipeline and it imports without a translator.

How to convert a render in Renovato

  1. Drop your hero render into the atlas as a source node. An exterior shot with the full building visible produces the cleanest mesh; partial views work but the occluded geometry is inferred and may be approximate.
  2. Drag the G.01 Image-to-3D preset onto the canvas and connect the source.
  3. In the inspector, pick the export format (GLB / FBX / USDZ) and the polycount tier (Lite ~50k tris, Standard ~120k, High ~200k). High polycounts capture more facade detail but produce larger files.
  4. Click Generate. Cost: 12 credits. Wall-clock: 90-180 seconds. The output node carries a download URL plus a built-in 3D preview that pans / orbits / zooms.

What it's good for

  • AR client previews — USDZ in iMessage lets a client place the building, life-sized, on their desk via iOS Quick Look. Zero setup on either end.
  • Web 3D embeds — drop the GLB into three.js or model-viewer for an interactive embed in a project page. Spinable model in the browser, no plugin.
  • Game engine handoff — Unreal and Unity import the GLB directly with PBR materials preserved. Use as a base shell to build the playable walkthrough on top of.
  • Variation studies — generate four variant images of a facade (M.01 brick, M.02 limestone, etc.), convert each to 3D, compare meshes side-by-side in the atlas.

What it's not for

Image-to-3D is a visualization output, not a BIM or CAD-grade asset. The geometry isn't metric — wall thicknesses, opening dimensions, and structural members are inferred from visual cues, not measured. Don't use the output for fabrication, energy modelling, or IFC-based handoff.

It's also not photogrammetry. Photogrammetry reconstructs accurate geometry from many overlapping photographs of an existing site; image-to-3D infers plausible geometry from a single render. For site context (terrain, neighbouring buildings), use photogrammetric capture and composite the AI mesh onto it. Renovato lets you do this — drop both as nodes in the atlas, treat them as separate sources for downstream presets.

Engine + format choices

Pick the export format by where the asset is going next:

  • GLB — Three.js, Blender, Unreal, Unity, model-viewer, Android AR. Default; ship this unless told otherwise.
  • USDZ — iOS AR (Quick Look, iMessage preview). Ship this for iPhone clients.
  • FBX — legacy 3D pipelines (older Autodesk, Maya, 3ds Max). Ship this when explicitly requested.

All three carry the same baked PBR maps — pick the container that fits the destination.

Frequently asked

What is AI image-to-3D for architecture?+
AI image-to-3D for architecture is a class of generative system that infers a plausible 3D mesh plus baked PBR materials from a single architectural render. The output is exported as GLB, FBX, or USDZ — usable in Unreal, Unity, Blender, three.js, model-viewer, and Apple AR Quick Look — without going back to the original 3D model.
Is the output BIM-grade?+
No. The geometry is a visualization-grade inference, not a metric capture — wall thicknesses, structural members, and opening dimensions aren't measured. Image-to-3D is for AR previews, web embeds, game-engine handoff, and variation studies, not for fabrication, energy modelling, or IFC handoff.
How is image-to-3D different from photogrammetry?+
Photogrammetry reconstructs accurate geometry from many overlapping photographs of an existing site; image-to-3D infers geometry from a single render. Photogrammetry is metric and slow (hundreds of photos required); image-to-3D is fast (one render, one preset run, ~2 minutes) but inferred. They serve different needs and complement each other.
What polycount and texture resolution do I get?+
Three tiers in the inspector: Lite (~50k triangles, 1k textures), Standard (~120k triangles, 2k textures), High (~200k triangles, 4k textures). PBR maps are albedo, roughness, metallic, and normal — bundled into the export so the model behaves correctly without manual material setup downstream.
How much does an image-to-3D run cost?+
Twelve credits per run on the standard tier. The Pro plan ($49/mo, 4,000 credits) supports ~330 image-to-3D runs per month. Wall-clock time is 90-180 seconds for the generation.
Which engine handles the 3D inference?+
Renovato routes G.01 to vendor-specific image-to-3D engines that change as the field matures — currently a mix of Hyper3D and Meshy-class models depending on scene type. The user-facing preset stays G.01; the underlying engine stays current.

← Back to blog

ch. 09 · begin

Drop a render.
See every version.

Open the studio, drop a render, watch Renovato relight, repopulate, animate and rebuild it. 60 free credits to start. No card.