Upload a BPMN image and select a vision model to extract structured output. Currenty supports only GPT-4o.