Helps non-vision models extract and understand images and screenshots via OpenRouter multimodal models.