API reference
HuggingFaceProvider Class
HuggingFace cloud provider – supports chat/VLM (IChatTask) and object detection (IObjectDetectionTask).
Normalized detections shape: [{ "score":float, "label":string, "box":[xmin,ymin,xmax,ymax] }, ...]
Properties
bool SupportsVision[Get]
Indicates whether this provider prepares and sends image inputs for supported multimodal models.
Disable for pure-text models to reduce request overhead and validation checks.See IChatTask and your model's capabilities on the provider dashboard.
override InferenceType DefaultSupportedTypes[Get]
string IUsesCredential. ProviderId[Get]
Unique identifier for this provider type (e.g., "OpenAI", "LlamaApi").
Used to store and retrieve credentials from the central CredentialStorage.
bool IUsesCredential. OverrideApiKey[Get]
When true, this provider asset uses its own API key instead of the central storage.
Member Functions
async Task< ChatResponse > ChatAsync
( ChatRequest req,
IProgress< ChatDelta > stream,
CancellationToken ct )
Sends a chat turn to a Hugging Face Inference endpoint and returns assistant text.
Supports optional image inputs when the selected model is multimodal.See IChatTask. HF Inference: https://huggingface.co/docs/api-inference/index
Parameters
req
Message payload and (optionally) images prepared by the editor or runtime.
stream
Optional progress receiver for partial tokens; final text is always delivered.
ct
Cancellation token for request assembly and HTTP.
Returns
ChatResponse with assistant text and raw payload for diagnostics.
async Task< string > DetectAsync
( byte[] imageJpgOrPng,
CancellationToken ct )
Runs object detection using a Hugging Face model and returns structured detections.
Handles tensor or JSON outputs depending on the model’s task definition.See IObjectDetectionTask. HF Tasks: https://huggingface.co/docs/inference-providers/tasks/object-detection
Parameters
imageJpgOrPng
Encoded image data (e.g., JPG/PNG). Must be non-empty.
ct
Cancellation token to abort pre-processing or HTTP.
Returns
Normalized detection set (boxes, scores, class ids) for downstream visualization.
async Task< byte[]> DetectAsync
( RenderTexture src,
CancellationToken ct )
Performs object detection on a RenderTexture by encoding it to JPEG and calling the byte[] overload.
This provides compatibility with the IObjectDetectionTask interface while adapting to cloud providers.
Parameters
src
Source RenderTexture to analyze.
ct
Cancellation token for aborting the operation.
Returns
UTF-8 encoded JSON string as byte array containing detection results.
Virtual reality / Meta Horizon
Support and legal
Virtual reality / Meta Horizon
Support and legal
Virtual reality / Meta Horizon
Support and legal
English (US)
© 2026 Meta