AI-103: Developing AI Apps and Agents on Azure

The input_fidelity parameter on an image edit request controls how closely the generated output preserves the visual details, identity, and features of the input image; setting it to high is intended precisely for edits involving faces, logos, or products where those details must be maintained in the output, at the cost of consuming more image input tokens than the default. Including a prompt and input image is simply what any image-edit call requires and does not by itself control identity preservation, a groundedness detection filter checks textual claims against source content (unrelated to image fidelity), and lowering temperature affects sampling randomness rather than image-edit fidelity.