Phi Ground Any 4B
Phi Ground Any 4B is a specialized screen-grounding model derived from Phi-3.5 Vision Instruct. Its integration depends on precise image preprocessing and prompt order.
What distinguishes this release
The official card specifies a fixed 1680 × 1008 input canvas, instruction-first formatting and a required system prompt. It supplies image-resizing and coordinate-conversion examples for its grounding output.Phi Ground Any 4B official model card
Build the task around its input and output
Practical guidance. Keep the scale and padding transform alongside each screenshot. Convert the predicted location back through that same transform before using it. A point that is correct on the model's canvas will not necessarily match the original application window.
Review the result in its intended setting
Practical guidance. Use the model to locate a visible target and let the application decide whether the action is permitted. After navigation, obtain a fresh screenshot. This is a grounding component, so it still needs surrounding control logic, state checks and recovery behavior.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | microsoft/Phi-Ground-Any |
| Model size | 4B vision modelPhi Ground Any 4B official model card |
| Input and output | Image and instruction input; grounded element outputPhi Ground Any 4B official model card |
| Weight license | MITPhi Ground Any 4B official model card |