NEWManaged multimodal retrieval.Explore platform →
    Models/Captioning/TencentARC/TimeLens-8B
    HFScene CaptioningOther

    TimeLens-8B

    by TencentARC

    8B video grounding model for timestamp-aware video understanding

    968dl/month
    8Bparams
    Identifiers
    Model ID
    TencentARC/TimeLens-8B
    Feature URI
    mixpeek://video_extractor@v1/tencentarc_timelens_8b_v1

    Overview

    TimeLens-8B is a video-text model from TencentARC focused on temporal grounding and video understanding. It is fine-tuned from Qwen3-VL-8B-Instruct on TimeLens datasets, making it relevant for finding when something happens, not just what appears in a sampled frame.

    On Mixpeek, TimeLens helps agents retrieve the right moment from long clips before taking action, summarizing evidence, or launching a follow-up workflow.

    Architecture

    Qwen3-VL-8B-Instruct fine-tune for video grounding. The model card lists TimeLens-100K and TimeLens-Bench datasets and video-text-to-text support.

    Mixpeek SDK Integration

    import { Mixpeek } from "mixpeek";
    const mx = new Mixpeek({ apiKey: "API_KEY" });
    await mx.collections.ingest({
    collection_id: "long-video",
    source: { url: "https://example.com/training-session.mp4" },
    feature_extractors: [{
    feature: "scene_caption",
    model: "TencentARC/TimeLens-8B"
    }]
    });

    Capabilities

    • Temporal grounding for natural language video queries
    • Video-text understanding
    • Timestamp-aware evidence retrieval
    • Built on Qwen3-VL-8B-Instruct

    Use Cases on Mixpeek

    Find the exact moment a product defect appears in inspection footage
    Retrieve video evidence for an agent before summarization
    Locate training, surveillance, or meeting moments from natural language

    Specification

    FrameworkHF
    OrganizationTencentARC
    FeatureScene Captioning
    Outputtext
    Modalitiesvideo, image
    RetrieverSemantic Search
    Parameters8B
    LicenseOther
    Downloads/mo968

    Research Paper

    TimeLens-8B

    arxiv.org

    Build a pipeline with TimeLens-8B

    Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

    Open Studio