Viewport vision

Kanzi AI Assistant can see the rendered result of its work. Instead of assuming that a successful tool call produced the right pixels, it can capture the Kanzi Studio preview and check what is actually on screen — a run-and-see loop that lets it confirm visual results such as shadows, materials, colors, and layout, and correct them if they are wrong.

You do not invoke these capabilities directly. Kanzi AI Assistant uses them on its own after building anything visual; this page explains what happens so you understand what you see in the chat.

Capturing the viewport

Kanzi AI Assistant captures the actual rendered output of the Kanzi Studio preview as an image. The capture renders offscreen, so a live preview window does not need to be open. Kanzi AI Assistant uses the captured image to verify a result before reporting success.

Describing the render

Kanzi AI Assistant can also have its configured model describe what the captured image actually shows, closing the run-and-see loop. This runs through whichever LLM provider you have selected in Settings:

  • If your selected model supports vision (image input), Kanzi AI Assistant sends it the render and reads back a description of what is really there.

  • If your selected model is text-only, Kanzi AI Assistant falls back to saving the render and asking you to confirm the result.

To get automatic visual verification, select a vision-capable model for your provider in Settings.

Importing images

Kanzi AI Assistant can bring an external image into your project:

  • Kanzi AI Assistant imports a standard image (PNG, JPEG, TGA, BMP) as a texture.

  • Kanzi AI Assistant imports an HDR image (.hdr, .exr) as image-based-lighting cubemaps, optionally including an environment background for image-based lighting.

Ask Kanzi AI Assistant to import an image and it chooses the right mode for the file you give it.