An agent that fires off edits but never sees the result is flying blind. v1.2’s vision loop fixes that.

viewport.screenshot captures the current frame; editor.read_image (and read_latest_screenshot) hand it back to the agent as an inline image. The model actually looks at the render.

See, then act

Frame the selection, take a screenshot, and tell me if the key light
is too harsh — if so, soften it and shoot again.

That’s a real loop: capture, evaluate, adjust, re-capture. Lighting, composition, material tweaks, “did the prop actually spawn where I meant” — all things that need eyes, not just return codes.

Why it matters

Vision turns Webified Bridge from a remote control into a collaborator. The agent isn’t guessing whether its change landed — it can see that it did.

Try it

Get Webified Bridge on Fab · Getting Started guide · Join the Discord