I wonder why the prompt seems generic in get_parsed_content_icon() , wouldn't that result in potential hallucinations? https://github.com/microsoft/OmniParser/blob/b0d5c9f5701f7e2be4771872e6e928da77759df3/util/utils.py#L102 if not prompt: if 'florence' in model.config.name_or_path: prompt = "<CAPTION>" else: prompt = "The image shows"