Ask the Video Editor assistant to analyze one photo, choose the objects to reconstruct, review the exact per-object generation quote, and then build a layered 3D scene with the photo as its backdrop.
Where: Native Video Editor → AI assistant → analyze_compositor_scene_image, then build_compositor_scene_from_image.
Key ideas
- Analyze first; generate only after approval: Photo analysis is a free planning step that reports object choices and an exact quote. Scene building may make paid model requests and must wait for explicit approval of that quote.
- Objects become editable scene layers: The chosen objects are generated with model choices per object and arranged with a camera/light plan; the original photo remains available as the scene backdrop.
- The scene joins a shared World: Generated layers enter the project’s World/scene composition so the user can inspect and continue editing the objects, camera, and lighting.
Steps
- Open a native Video Editor project and make sure the source photograph is available in the project media.
- Call analyze_compositor_scene_image with that photo. Review the detected objects and exact total quote; analysis alone does not generate objects.
- Choose which detected objects to include or omit and confirm model/quality choices where offered.
- Only after the user explicitly approves the quoted Sparks cost, call build_compositor_scene_from_image with the accepted analysis and selections.
- Inspect the resulting World, generated object layers, backdrop, camera, and lighting. Adjust placement and materials before rendering.
- If geometry/textures are too heavy for preview or render, use Scene Model Optimizer as a separate reversible-by-undo step and compare the result.
AI tools and cost gate
- analyze_compositor_scene_image performs the free analysis and returns object choices plus the exact quote.
- build_compositor_scene_from_image creates the selected model layers and incurs the quoted generation cost.
- Do not start build_compositor_scene_from_image before the user approves the exact current quote; analysis is not consent to spend.
What to review after generation
- Generated objects are not guaranteed to match the source photograph pixel-for-pixel; inspect silhouette, scale, occlusion, material, and camera alignment.
- The source image backdrop helps preserve visual context while the reconstructed geometry remains editable.
- The Flowgen Scene3D node offers a related scene-from-image path with its own graph contract; this article describes the native compositor tools.
Tips
- Begin with a clear photo with separated objects and a visible ground plane; simpler scene structure generally makes the first reconstruction easier to inspect.
Limitations and important notes
- Object detection and generated geometry are approximations, not CAD reconstruction.
- The build step may incur Sparks cost and requires explicit approval of the current exact quote.