NEWVectors or files. Pick a path.Start →
    Runs as a custom pluginEmbeddingRecommendedVerified 2026-11-05

    SigLIP 2 so400m patch14 (224/384)

    google/siglip2-so400m-patch14-384

    SigLIP 2 so400m patch14 (224/384) is a image and text embedding model (Apache-2.0). 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.

    Record as JSON

    No demo for SigLIP 2 so400m patch14 (224/384) 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
    Logos, signage, brands and venue shots in concert footage (Live Nation visual arm); strongest SigLIP 2 below giant.
    Worst at
    Alone it trails text retrieval on speech and moment queries (Live Nation visual F1@10 0.348 vs text 0.414); 1.1B params, GPU in practice.
    Task
    Embedding
    Input
    Image, Text
    Output
    Vector
    Use cases
    Visual search, Brand detection, Multilingual search
    Why we recommend it
    Apache-2.0, safetensors, no remote code; carried brand and venue queries in the Live Nation held-out eval.

    Templates that use it

    No template uses this model yet. Browse templates.

    Input and output

    Input

    image
    resolution
    384x384 (224 sibling)
    color
    RGB
    normalize
    mean 0.5, std 0.5
    text
    max_tokens
    64
    prompts
    lowercase captions; the card uses 'this is a photo of {label}.'; Gemma tokenizer, padded to max_length 64

    Output

    vector
    dims
    1152
    distance
    cosine
    type
    dense
    note
    image and text vectors share one space; L2-normalize before cosine

    How Mixpeek wires it

    Extractors
    • ln_shots@v1 fixed (customer plugin, customers/livenation/manifest-loop)
    Feature URIs
    • mixpeek://ln_shots@v1/ln_visual
    Vector indexes
    ln_visual
    Retriever stages
    feature_search
    Combines with
    • shots -> one to three frames per shot -> SigLIP 2 vectors (mean) -> feature_search
    • SigLIP 2 visual + Qwen3-Embedding text + BM25 on search_text -> DBSF fusion (Live Nation tri retriever)

    How it runs

    Status
    Runs as a custom plugin Runs today as the Live Nation customer plugin ln_shots@v1 in the standalone image (safetensors, local_files_only, trust_remote_code false); not a built-in Serve app.
    Serve apps
    None
    Runtime
    transformers AutoModel + AutoProcessor (Siglip2Model)
    Weights
    safetensors, no remote code
    Parameters
    1,136 M
    Size on disk
    4,544 MB
    Preprocessing
    Resize to 384x384, rescale to 0..1, normalize mean 0.5 std 0.5; text padded to 64 tokens (max_length).
    Postprocessing
    L2-normalize pooled image and text embeddings; sigmoid of scaled logits for zero-shot scores.
    Serving
    both, batch 32
    Hardware
    Runs on CPU (slower without a GPU)

    Scale to zero

    Minimum replicas
    0
    Maximum replicas
    1
    Idle timeout
    600 s
    Memory
    8 GB
    CPUs
    2
    GPUs
    1
    Cold start, node warm
    Not recorded
    Cold start, new node
    Not recorded

    Measured cost

    Unmeasured

    No committed measurement; plan row 3 measures it in the standalone image.

    Our evaluations

    Dataset and metricScoreBaselineDate
    Live Nation marketing clips, 10 clips, 263 shots, 56 queries
    F1@10, visual arm alone (k=10)
    customers/livenation/manifest-loop/README.md
    		0.348text arm alone (Qwen3-Embedding-0.6B): 0.4142026-11-04
    Live Nation held-out clips, 5 clips, 28 queries
    F1@10, SYSTEM: tri retriever (SigLIP 2 visual + Qwen3 text + BM25, DBSF) (k=10), held out
    customers/livenation/manifest-loop/DRIVER-RUN.md
    		0.406None2026-11-04

    Questions

    What is SigLIP 2 so400m patch14 (224/384) best at?
    Logos, signage, brands and venue shots in concert footage (Live Nation visual arm); strongest SigLIP 2 below giant.
    What are the limits of SigLIP 2 so400m patch14 (224/384)?
    Alone it trails text retrieval on speech and moment queries (Live Nation visual F1@10 0.348 vs text 0.414); 1.1B params, GPU in practice.
    Can SigLIP 2 so400m patch14 (224/384) be used commercially?
    Yes. Apache-2.0 weights and code; no use restrictions.
    How does SigLIP 2 so400m patch14 (224/384) run on Mixpeek?
    Runs today as the Live Nation customer plugin ln_shots@v1 in the standalone image (safetensors, local_files_only, trust_remote_code false); not a built-in Serve app.
    What does SigLIP 2 so400m patch14 (224/384) output?
    A 1152-dimensional dense vector, compared by cosine similarity.