Runs as a custom pluginFaceVerified 2026-11-16
AuraFace v1 (SCRFD-10GF + ArcFace-style ResNet100)
fal/AuraFace-v1
AuraFace v1 is a face recognition model (Apache-2.0) for commercial face search. Can run on Mixpeek as a custom plugin.
Mixpeek does not run it today. It can run as a custom plugin: the weights load without remote code.
No demo for AuraFace v1 (SCRFD-10GF + ArcFace-style ResNet100) yet
A video, image, slideshow or playground goes here once we record one on our own data.
What it is good and bad at
- Best at
- 512-d face identity vectors under a commercial license, loaded the same way as an InsightFace pack (FaceAnalysis with the pack's directory), so it drops in where buffalo_l runs.
- Worst at
- Separating a stranger from a match: on the face-check benchmark a person not in the index scored up to 0.418 against a true-match median of 0.55 (buffalo_l: 0.271 against 0.613), and the model card notes its training data may not cover every ethnicity well.
- Task
- Face
- Input
- Image, Video
- Output
- Boxes, Keypoints, Vector
- Use cases
- Face search
Templates that use it
No template uses this model yet. Browse templates.
Input and output
Input
Image
- Det size
- 640x640
- Align
- 5-point affine to 112x112
Video
frames sampled by the extractor
Output
Boxes
face bounding boxes with detection scores
Keypoints
5 landmarks per face
Vector
- Dims
- 512
- Distance
- cosine
- Normalized
- true
How Mixpeek wires it
- Extractors
- None today
- Feature URIs
- None today
- Vector indexes
- None
- Retriever stages
- feature_search
- Combines with
- image or video -> SCRFD boxes -> aligned 112x112 crops -> AuraFace 512-d vector per face -> feature_search with a face photo
How it runs
- Status
- Runs as a custom plugin No Serve app or built-in extractor loads it yet. face_identity_extractor loads the buffalo_l pack by name; AuraFace needs the pack name (or a custom plugin) pointed at the AuraFace directory.
- Serve apps
- None
- Runtime
- insightface.app.FaceAnalysis(name="auraface", allowed_modules=["detection", "recognition"]) on onnxruntime, det_size=(640, 640)
- Weights
- ONNX, no remote code
- Parameters
- Not recorded
- Size on disk
- 265 MB
- Preprocessing
- SCRFD detection at 640x640, 5-point affine alignment to 112x112.
- Postprocessing
- Recognition embedding per aligned face (512 dims), L2-normalized.
- Serving
- both
- Hardware
- Runs on CPU (slower without a GPU)
Scale to zero
- Minimum replicas
- Not recorded
- Maximum replicas
- Not recorded
- Idle timeout
- Not recorded
- Memory
- Not recorded
- CPUs
- Not recorded
- GPUs
- Not recorded
- Cold start, node warm
- Not recorded
- Cold start, new node
- Not recorded
Measured cost
Unmeasured
No committed throughput or cost measurement.
Our evaluations
| Dataset and metric | Score | Baseline | Date |
|---|---|---|---|
ip-check face-check benchmark, 82 photos of 4,978 indexed public figures (a different photo from the indexed portrait) recall@1, largest face per image, offline (k=1), held out | 0.927 | insightface-buffalo-l on the same photos and index: 0.976 | 2026-11-16 |
Questions
- What is AuraFace v1 best at?
- 512-d face identity vectors under a commercial license, loaded the same way as an InsightFace pack, so it drops in where buffalo_l runs.
- What are the limits of AuraFace v1?
- It separates strangers from matches less cleanly than buffalo_l (a stranger scored up to 0.418 against a true-match median of 0.55), and its training data may not cover every ethnicity well.
- Can AuraFace v1 be used commercially?
- Yes. The recognition weights are Apache-2.0 and trained for commercial use; the bundled SCRFD detector comes from InsightFace's model zoo, so replace it before relying on the pack commercially.
- How does AuraFace v1 run on Mixpeek?
- As a custom plugin today; no Serve app or built-in extractor loads it yet.
- What does AuraFace v1 output?
- A 512-dimensional face vector per detected face, compared by cosine similarity.