Multimodal Runtime Inference

Reusable execution engines primarily specialized for image, audio, video, vision-language, or combined multimodal models. Complete-engine deployment and hardware precedence defined by llm-inference applies first. Excludes model releases, training tools, end-user media workflows, and general engines with incidental multimodal support.