Enter a task description and see where in the image affords that action. Note that the model works best for action queries rather than object semantic properties like colors. Check out the examples or try with your own image!