GenAnywhere Generate Now

Select a map area. Generate a 3D city model with AI.

No drone survey required. Turn map data into editable 3D buildings, terrain, water and vegetation for BIM, GIS and 3D workflows.

Generate now

VR game prototypes: a 3D scene checklist GenAnywhere supports

A practical checklist for the site scene behind a VR prototype, from the opening boundary to the handover note.

A VR game prototype needs ground to stand on, and GenAnywhere can supply that ground as an editable site model generated from a map area. The checklist below covers what such a scene should hold, what it should prove and what it leaves to the engine team. None of it is a level. All of it can stop a prototype from being built on the wrong assumptions.

Warehouse district street with rail sidings and low industrial buildings

1 Fix the question the prototype answers

1.1 What decision should the prototype settle?

Write the decision at the top of the page. A prototype asking whether a street is wide enough for two players at once needs different content from one asking whether a courtyard reads as a courtyard from the entrance. Pick one, and let the others wait for the next build.

One decision per build also keeps the session short. When testers are asked three questions, they answer the easiest one and guess at the rest.

Write the decision in the same words every time it comes up. A prototype discussed as "does the market square work" in January and "is the square big enough" in March has quietly become two projects, and nobody notices until the answers stop matching the questions.

1.2 How much of the site does the prototype need?

Draw the area the decision touches, plus a margin. GenAnywhere runs AI 3D generation on the boundary you set, producing buildings, terrain, water and vegetation inside it, and the margin matters because a prototype judged from inside a tight box answers the wrong question. Buildings can be replaced, moved or adjusted in height before export, which is how one site can serve two test variants without a second drawing.

The margin does not have to be detailed. It has to be present, so the player understands there is a world beyond the test area and stops trying to walk out of it.

2 The checklist before anyone scripts

2.1 What belongs in the scene at day one?

Ground, massing, boundary and a human scale reference. Ground because everything sits on it, massing because it controls sightlines, and a boundary because a prototype without edges invites the player to walk into geometry that was never built. Add water if it affects movement. That is a scene someone can review in a headset within a session.

Then add a marker for the play area and a marker for the spawn. Two coloured planes on the ground do more for an opening read than a week of props.

The generated scene arrives with buildings, terrain and water already separated, which is why the checklist can treat them as layers rather than as one mesh. Hide what you are not testing and the session gets simpler.

2.2 Which detail can wait?

All of it, until the detail changes a choice. A doorway too narrow for the player's collision capsule matters. A door handle does not. Ask testers what they noticed, then add the items that came up twice.

Waiting is cheaper than removing. A prototype with too much content buries the question it was built to answer, and the team spends the session admiring a facade instead of measuring a street.

3 Entry, sightlines and scale

3.1 Where does the player start and what do they see?

Put the spawn where the real arrival would happen: a station exit, a car park, a gate. The opening seconds carry more information than the rest of the session, because the player is still calibrating what kind of place this is. Stand there yourself before asking anyone else to.

Walk the entry route twice, once at the pace of someone arriving and once at the pace of someone who already knows the place. The second walk is usually faster and skips the hesitation that made the first one useful. Note where a tester stopped without being asked, since a pause is often the first sign that the scene is asking a question the design has not answered. Two walks also expose a route that only works when the player already knows where to go. None of this needs a script. It needs someone to watch quietly and write down the time.

Then check the sightline the design depends on. If the landmark is supposed to be visible from the entrance, it either is or it is not, and no plan view settles that.

Reading order matters as much as the view itself. A player who sees a wall, then a gap, then a lit space has been told where to go without a word of instruction, and the blockout is where that sequence gets tested cheaply.

3.2 How do you test scale in the headset?

Use a capsule or a simple avatar at the intended height, walk a measured distance, and count the steps. Repeat the same distance in the headset. Distances read shorter in VR, and a square that looks generous on a plan often feels cramped from eye height.

Record both numbers. The gap between them is the correction the art team will need later, and it is easier to defend than an opinion.

Scale model of a dense block beside a waterfront promenade

4 Interaction and comfort belong downstream

4.1 Which interactions must be authored by hand?

GenAnywhere supplies geometry, not interaction. Doors, levers, handholds, vehicles, throwables and menus are built in the engine, and a site model carries no rig and no animation, so adding any of those is project work. Plan the prototype around one interaction at most, so the scene keeps its job as a spatial test.

Keep the interaction attached to a placeholder object. When the placeholder is replaced, the question of whether the space works stays answered.

4.2 When should comfort testing begin?

As soon as there is movement, even if the movement is a white capsule gliding along a street. Slopes, uneven ground and turns are the three things most likely to produce discomfort, and all three exist in a site blockout before any art does.

Test with one person who is sensitive to motion and one who is not. The pair tells you more than an average, and it takes ten minutes.

Write the comfort notes down while the headset is still on someone's head. Recollections an hour later tend to soften, and a mild complaint is often the useful signal.

5 From prototype scene to production level

5.1 What survives the handover?

The site geometry and its arrangement. GenAnywhere exports a general GLB carrying buildings, terrain, water and vegetation on separate layers, previewed before download, and that file becomes the reference the production level is measured against. The platform is clear about what it supplies: a site model built from map data, with estimated heights. What a headset does with that geometry belongs to the engine.

Everything else in the prototype is scaffolding, including the spawn point, the capsule and the lighting rig, and none of it should be mourned.

5.2 What gets replaced?

Walls the player touches, floors they stand on and openings they pass through. A generated facade carries the right mass and the wrong detail, and the close-up view is where that shows. Replace the facade, keep the footprint, and check the replacement against the blockout before deleting anything. The GLB follows the glTF specification, an open format for 3D scenes, which is why the reference usually opens in whatever tool the art team prefers.

Artists who can compare against the blockout make faster decisions than artists working from a photograph and a memory.

Aerial view of a low campus beside a river and forest

6 Run the checklist before the build grows

6.1 Who should walk the prototype?

Someone from design, someone from art and one person who has never seen the project. Add anyone who knows the real site, if the prototype is based on one, because they catch the errors that look convincing to everyone else. Four people is enough for an opening session.

6.2 What counts as a passed check?

A check passes when the team can state the answer in one sentence and point to the moment in the session when it appeared. "The street is too narrow for two players" is a pass. "It felt off" is a prompt for another question, and the next session should be built around it.

Write the answer on the checklist before the room clears. A pass that lives in memory alone turns into a debate two weeks later, when nobody can remember who said it or what they saw.

6.3 When do you stop adding to the prototype?

When the decision is answered, or when the next question needs a different scene. Prototypes age badly, because every addition makes them harder to read. Keep the build small enough that a new tester understands it in a minute, and start a clean scene when that stops being true.

Archive the old prototype rather than deleting it. The next team inherits a record of what was tried, and the reason a street was rejected is often more useful than the street itself.

Draw the prototype site in the GenAnywhere app and walk the entry route before the next playtest.

7 FAQ

FAQ-1 Does GenAnywhere build a prototype level?

No. It builds an editable site model from a drawn map area and exports a general GLB, so the geometry can be imported into an engine and used as the spatial base of a prototype. Collision, interaction, lighting, audio and performance work remain engine tasks. The value is the arrangement: real building masses, terrain and streets in place before anyone models a hero asset.

FAQ-2 How large an area should we draw?

Start with the play area plus one street or block of margin. A larger boundary gives better sightlines and a heavier file, and the prototype rarely needs both at once. If the decision is about a single junction, draw the junction and enough of the approaches to judge it. Expand the boundary later if the second question needs it, and note the change so the two sessions can be compared.

FAQ-3 Do we need final art before testing?

No, and waiting for it is the usual way a prototype slips. Grey massing, simple ground and a human scale reference answer questions about width, distance and sightlines. Art answers a different question, which is whether the place feels like itself. Run the spatial test before the art pass, since a beautiful level built on a cramped street stays cramped.

FAQ-4 How should we record what testers noticed?

One page per session: the decision, the scene version, who was in the headset, what they said and what it implies for the next build. Keep it beside the site file rather than in a chat thread. Two sessions later, the page is the reason a decision is on record, and the person who missed the test can still follow the argument.

See how a drawn area becomes an exported file in the GenAnywhere workflow overview.

Content and technical notes

This article was prepared by the GenAnywhere team with AI-assisted drafting and human review.

Product features, formats, and availability change as the platform evolves; refer to the official GenAnywhere website for current details.

For technical questions or corrections, contact support@genanywhere.com.

Last reviewed: 2026-11-06.