Interactive Offline Vision LLM

Interactive Offline Vision LLM is a package that will let you ask natural-language questions about static images, RenderTextures, video frames and live camera feeds using Liquid AI's LFM2.5-VL-450M model. It runs completely offline after the initial model download, without any networking or services required at runtime.

It does not contain platform-specific code or native libraries, so it can be used on desktop, Android and iOS platforms supported by Unity AI Inference. Inference can be performed on the CPU or GPU Compute backend. WebGPU inference is experimental and is not officially supported in this release; current browser performance and stability are not considered production-ready.

A ready-to-use InteractiveOfflineVisionLLM prefab provides clean RunPrompt and Ask APIs, with streaming callbacks, stop control, status events and a selectable backend. An included DemoScene gets you up and running quickly, while the internal VisionInference class keeps Unity AI Inference workers, tensors and cache implementation behind the customer-facing component.

The model is not included with the package. The Inspector downloader opens Liquid AI's licence and requires acknowledgement before download. The publisher-provided Unity conversion is downloaded into Assets/Models; only this initial Editor download uses networking, and inference is local afterwards.

The optional LFM2.5-VL-450M model is downloaded separately under the LFM Open Model License v1.0 and is not included in this Asset Store package; see Third-Party Notices.txt in the package for details.