Workflow

Image to 3D assets for Unity

Atlas reads one reference photo and rebuilds it as five modular, engine-ready 3D kit pieces — wall, corner, floor, doorway and trim — all locked to the same materials and palette.

Try it on Atlas

Four stages, one photo to a modular kit

01A stone corridor reference photo with an arched doorway, a hanging lantern and a carved wall sconce

Visual deconstruction

A vision LLM inspects the reference photo’s architectural grammar — wall patterns, joints, pillar types, floor surfaces, portals and props — and outputs a numbered list of 5 isolation prompts.

02

Prompt fan-out

Five extractor LLMs parse the master list in parallel, each taking one numbered slot and emitting a clean isolation prompt.

03Four isolated concept pieces from the kit — an arched doorway, a column, a floor tile and a wall panel with a hanging lantern

Reference-locked synthesis

Five multimodal nodes generate the individual modular pieces in parallel, locking style, texture and palette to the source photo while isolating each piece on a solid gray background.

04The same four kit pieces as shaded 3D meshes, ready for grid-snap assembly in an engine

Mesh & texture generation

Five Image-to-3D nodes reconstruct each concept into a watertight mesh with full PBR texture maps.

Where the time goes

Stage
Manual
Atlas
Vision analysis & prompt extraction
3–8 hours
5–10 seconds
2D concept isolation (parallel)
3–5 days
20–35 seconds
3D mesh & PBR texturing (parallel)
2–3 days
90–120 seconds
Total time for the kit
1–2 weeks
2–3.5 minutes

Common questions

What does this workflow output?
Five modular GLB assets — a straight wall segment, a corner pillar, a floor tile, a doorway arch and a wall trim piece — each with a full PBR material set, plus the 2K concept renders and a written architectural breakdown.
How long does it take?
About 2 to 3.5 minutes for the full kit. Because all five asset pipelines run concurrently, five modular pieces take almost the same time as generating one.
Can I control the art style?
Yes — the generation nodes reference your photo directly, so whatever aesthetic it carries (hand-painted, photorealistic, low-poly voxel, sci-fi metallic) is what comes out. You can also edit the master vision prompt to push a specific direction, or switch the face-count setting for higher fidelity or a lower-poly runtime budget.
Which engines are supported?
Standard GLB — native import in Unity 2022+ via its official importer package, Unreal Engine 5 via Datasmith, Godot 4 in one click, and Blender, Maya or 3ds Max, plus Three.js and Babylon.js for the web.

Related workflows

AI 3D environment generator →Sketch to 3D model with AI →AI rigged character generator →

Try this workflow in Atlas

Free to start. No credit card required.

Try it on Atlas