Scene props from a screenshot

What this lane does

You drag a box over the 3D viewport. iCity renders that region, uploads the picture, and the cloud looks at it and decides which props belong in that scene — the same era, region, scale, material language and time of day as the buildings already in shot. It then edits your own screenshot in place, adding those props to it, and that edited picture is what you approve.

Nothing is searched in the library and nothing is placed in 3D. The props are invented for this view. What comes back is one image per approved prop, saved into your project folder. If you want assets that enter the library and get placed on the city, that is a different lane — Single materials and props and Generating a theme.

Before you start

• —You are signed in. The send resolves your account token first and refuses with not signed in before anything is uploaded.
• —A project is open. The finished prop images are written into the active project folder. With no project open they fall back to a temporary folder instead — see What a project is.
• —The view is framed the way you want it. The capture is taken from the viewport matrices at the moment you release the mouse, so the cloud reasons about exactly that camera.

The capture follows your viewport shading. In Material Preview or Rendered, and when the scene has a world and lights of its own, the render uses the scene's own engine so the picture matches what you are looking at. Otherwise it falls back to Solid/Workbench, because a real render of an unlit city comes back black.

Taking the capture

Everything happens in the Generate with AI popup, on its Screenshot tab. In the Design composer the quick row From a screenshot opens the popup already on that tab; otherwise open the popup and click the tab yourself.

With nothing captured yet, the footer button reads Capture area. The camera icon at the right end of the icon strip under the prompt box starts the same capture. Note that the Screenshot tab draws no attach or library-asset icons: your capture and your text are the whole request.

1. Click Capture area. The popup closes and the viewport enters the box modal — the status bar reads Select Box for drag and Cancel for Esc.

   scene-props-from-a-screenshot-drag-box.png

   The 3D viewport mid-drag with the capture rectangle drawn over a dressed street view, and the status bar at the bottom of the window showing 'Select Box' and 'Cancel'.

   /docs-shots/blender/

2. Drag a box over the region you want and release. Press Esc or right-click instead to back out; a cancel from the popup returns you to the popup and takes no shot.
3. The region renders and the SCREENSHOT review popup opens with the picture. The wheel zooms it and dragging pans it. Click Accept to keep the shot, or Retake to discard it and drag a new box over the same target.

   scene-props-from-a-screenshot-review-popup.png

   The GPU popup titled SCREENSHOT showing a freshly rendered viewport capture, with the Retake and Accept buttons at the bottom right and the close button in the header.

   /docs-shots/blender/

4. Accept returns you to the Screenshot tab with the shot on it as a small tile below the prompt box. The × on that tile drops the capture; the camera icon takes a new one.

Sending it

With a capture held, the footer button reads Generate. Type what you want added — the tab asks What to add to this area and suggests e.g. add garbage bags and a bicycle near the wall — then press it. Your text is optional here: with the box empty the cloud works from the picture alone, and when you do write something, your request wins wherever it disagrees with the model's reading of the scene.

The render is encoded and posted with the job, and the popup closes. If the send is refused the popup stays open so the shot is not lost, and the reason is written to the log. Progress is tracked like any other job — Jobs, progress and errors.

Screenshot sent — imagining props for your view (a1b2c3d4…)

What the cloud does with it

The picture arrives as the job's first reference image, through the same upload path every reference image in iCity uses. The lane then runs in two moves before it stops and waits for you.

1. The prop list. A vision pass over your screenshot proposes the props that belong in that view — street furniture, vegetation, signage, vehicles, everyday clutter. It is told to propose free-standing objects only: never buildings, roads, terrain, sky, or anything already in the picture. Four props by default.
2. The approval image. Your same screenshot is edited in place with those props added. That composite — not the list of names — is what you are shown, because approving a list of words is not approving a result.

If the composite fails to render, the run does not die: the plan is the durable half and the job can still take a decision on it. After you approve, each prop is rendered again on its own on neutral ground, conditioned on the approved composite rather than cropped out of it, so the scene's pavement, shadows and camera angle do not travel with the object.

Approving the props

When the job reaches its pause, the SCENE PROPS popup opens with the composite. The wheel zooms, dragging pans, and a double-click shows it across the whole region. One button, Approve, whose tooltip reads Have each of these props made as an image.

Closing the popup instead does not cancel anything — its close button says Close (the job keeps waiting for a decision), and the job stays parked in the cloud until it gets one. Job tracking is session-only, so a job left waiting while Blender is closed cannot be picked back up from the interface.

Where the images land

After approval the per-prop renders come back, are downloaded, and are saved before you see them — a popup you close should not be the only copy. The Scene Props results popup then shows the finished thumbnails as a contact sheet, with the count, the folder they were written to, and each prop's name under its picture.

The folder is screenshot_props inside the active project, with one subfolder per job named from the first eight characters of the job id, and one PNG per prop named after the prop.

<project folder>\screenshot_props\a1b2c3d4\Folding bicycle.png

These images are the deliverable. They are not library assets: they do not appear in the asset browser, they are not part of your theme, and nothing is added to the 3D scene. Placing them would need meshes, and the mesh half of this lane is not switched on.

Cost and limits

The popup's footer shows Est. cost: ~11 credits (est.), which is the formula rather than a measurement: one planning call, one approval image, and one render per prop. Four props is the default count. For how credits work, see Credits and plans.