Skip to content
EN
Get API key

VIDEO DATA FOR AI & DATA TEAMS

Video Datasets for Multimodal AIReliable collection. Direct to your cloud.

Start with billions of video IDs or bring your own. We handle YouTube bot challenges, proxy routing and recovery to keep collection moving. Build with our APIs or agree a managed delivery of media, metadata and available captions.

YouTube video IDs available
Billions
Aggregate daily delivery capacity
1–10 PB
Video IDs per bulk submission
100M+
Cloud storage destination types
8

BUILT FOR YOUTUBE

Keep your pipeline moving through bot checks.

The same YouTube retrieval infrastructure powers our APIs and managed dataset delivery. We handle access challenges and recovery while your team focuses on its data.

Bot challenges handled

YouTube-specific bot detection and challenge handling work with preselected residential proxy routes. Your team does not need to source or operate the proxies.

Fast media retrieval

Individual video downloads can exceed 20 MB/s. Get media moving toward your editing, transcription or AI pipeline sooner.

Know the outcome

Managed retries and recovery support reliable downloads. Track job results, file locations and failures so your next processing step knows what arrived.

Evaluate your own workflow.

Check the output, delivery result and usage on representative videos before expanding your collection.

TWO WAYS TO GET YOUR DATA

Build it yourself. Or leave delivery to us.

Build your video data pipeline

Keep control of discovery, selection and orchestration. FlashData handles media retrieval, bot challenges and delivery so your team can focus on processing the data.

  • Discover videos and retrieve metadata with YouTube Data APIs.
  • Submit your ID files through the Bulk Download API.
  • Track individual outcomes and receive media in your bucket.
Explore the API workflow

Get a dataset delivered

Hand off collection and delivery to our team. Start with our IDs or your own list, evaluate a representative sample, and agree the outputs, pricing and delivery checks.

  • Tell us the collection, volume and media you need.
  • Evaluate a sample against your requirements.
  • Agree the scope and receive the dataset in your storage.
Request a dataset sample

WHAT YOU RECEIVE

See what your pipeline will receive.

Evaluate sample outputs and delivery records before expanding the collection. Media, metadata and existing caption tracks are separate deliverables, selected for your project.

Video & audio

Video with audio, video-only files or audio-only collections. Choose formats, quality and supported clip bounds.

Structured metadata

Video identifiers, titles, channels and other available source fields. Optional companion JSON keeps records alongside media.

Existing captions

Request available transcript or subtitle tracks as a separate deliverable. Track availability and language depend on the source.

Delivery records

Inspect per-video results, file locations and failures. Keep the records needed to continue your downstream pipeline.

Example video frame from the FlashData YouTube workflow
Example media assetFUfGcZ092b0.mp4
dataset/
  media/FUfGcZ092b0.mp4
  metadata/FUfGcZ092b0_meta.json
  delivery/manifest.json

Illustrative file layout, not a downloadable dataset sample. Actual fields and contents are agreed for your project.

PUBLIC DATASET DOWNLOADS

From public datasets to your bucket.

Bring YouTube IDs from LAION-BVD, HowTo100M and more. We download the available video or audio directly to your cloud storage.

FlashDataBulk download & delivery
Your cloud storageAmazon S3 · R2 · GCS · Azure
media/FUfGcZ092b0.mp4
metadata/FUfGcZ092b0_meta.json
delivery/manifest.json
Illustrative delivery layout
Dataset details & availability
LAION Big Video Dataset (LAION-BVD)1.3B URLs across platforms
Use the YouTube entries in BVD-URLs, its URL-only release. LAION-BVD is released for non-commercial research.
HowTo100M1.2M source videos
Instructional videos behind 136M annotated clips. Use the source YouTube IDs to retrieve available media.
HD-VILA-100M3.3M source videos
YouTube URLs and timestamps for 100M clips. Retrieve the source videos for your own clip-processing pipeline.
InternVid230M video-text pairs
Clip annotations include YouTube IDs, timestamps and captions. Many clips can refer to the same source video.
AudioSet2M+ annotated audio segments
YouTube IDs, timestamps and sound labels. Retrieve available video or audio for your own segment processing.
YouTube-8M6.1M video references
Video labels and audiovisual features. Resolve anonymized dataset IDs to available YouTube IDs before submission.

Published sizes describe the original releases, not current download availability. FlashData retrieves available YouTube source media, not a guaranteed replica of a dataset or access to restricted archives. Retain original annotations and timestamps for downstream processing. Dataset and source-video terms apply; no affiliation or endorsement is implied.

YOUR DATA. YOUR ENVIRONMENT.

From video IDs to your cloud storage.

Deliver to your existing bucket or container. Media access and retention stay under your storage policies.

  1. Video ID collection
  2. Sample & output selection
  3. Collection & delivery
  4. Your cloud storage
  • Amazon S3
  • S3-compatible / R2
  • Google Cloud Storage
  • Azure Blob
  • Alibaba OSS
  • Tencent COS
  • Baidu BOS
  • Volcengine TOS
View storage configuration

FOR DEVELOPERS

One pipeline, from discovery to delivery.

Use each API independently or connect them in your application. You control selection and orchestration; FlashData handles retrieval and media delivery.

  1. 01

    Find and enrich video IDs

    Search videos and retrieve metadata or available caption tracks.

    YouTube Data API
  2. 02

    Process your ID files

    Submit large collections with shared output settings. Track progress and pause or resume new submissions.

    Bulk YouTube video downloads
  3. 03

    Receive video and audio

    Choose media formats, quality and storage. Record each result for your downstream processing.

    YouTube Download API

Bulk video download — Python toolkit

YouTube Search API: Python Guide · YouTube Search API Python toolkit

Self-service API usage follows your plan limits and available Credits. Download charges settle against delivered media bytes; your storage provider's charges are separate.

See API pricing

COSTS THAT SCALE WITH YOUR WORKLOAD

Competitive rates. Pay by delivered media.

Start with pay as you go and discuss custom unit rates based on your usage volume, or choose a subscription with lower published download rates. Final usage follows the media bytes delivered; optional companion metadata JSON is excluded from download-volume charges.

These are self-service API rates. Managed dataset collection, samples, packaging and delivery are scoped and quoted separately.

Pay as you go

$100.00per TB delivered

No monthly subscription. Top up Credits as needed.

Volume pricing for pay as you go

Share your expected usage volume to discuss custom pay-as-you-go unit rates with our team.

Discuss volume pricing

Business

$40.00per TB delivered

$1,999.00 / month · subscription

Includes approximately 50 TB per billing period if all included Credits go to downloads.

Example: 100 GB delivered

$10.00 of Credits at the pay-as-you-go rate. File size determines download usage; video count alone does not.

Subscriptions include usage Credits shared across FlashData APIs. Usage covered by those Credits is not charged again. Your storage provider’s charges are separate. 1 TB = 1,000 GB.

BUILT AROUND YOUR WORKLOAD

Video data for multimodal workflows.

Multimodal model development

Collect the video, audio and available text inputs your team needs for its research and evaluation pipeline.

Video understanding & retrieval

Build collections for indexing, search, summarization and video analysis.

Audio & speech workflows

Receive audio-only media for your own transcription, processing or evaluation tools.

Data availability does not establish permission for model training or other uses. Scope, provenance and applicable rights need to be evaluated for your project. Existing captions are not newly generated transcriptions or translations.

START WITH A SAMPLE

Evaluate the data before the full delivery.

Tell us the sources, outputs and destination you need. Evaluate a representative sample for coverage and quality, then agree collection costs, delivery checks and timing. No storage credentials are needed for the initial request.

Dataset scope, pricing and delivery commitments are agreed separately from self-service API plans.

WHAT YOU RECEIVE

Before you start.

What types of data are included in the YouTube datasets?
Depending on your agreed scope, datasets can include video IDs, video or audio files, structured metadata and available transcript or subtitle tracks. Metadata can include titles, channels, publication dates, durations and engagement fields. Availability varies by video; not every record includes every field or caption track.
In what formats are the datasets delivered?
Video formats include MP4, WebM and MKV; audio formats include M4A, MP3 and Opus. Structured metadata, transcript or caption records and API results are available as JSON. The requested assets and packaging are confirmed before managed dataset delivery.
What is the quality of the video and audio content?
Choose video resolution, frame rate and audio-quality preferences, including 4K or 8K where matching source formats are available. Actual quality depends on the source and selected output; requesting a higher resolution cannot improve the original content. Evaluate a sample before scaling, since unavailable quality or format combinations can fail.
Is the data suitable for model training?
Video, audio, metadata and available captions can support multimodal model development, training and evaluation. Evaluate a sample for the coverage, quality and alignment your pipeline needs. Data availability and source license labels do not by themselves establish training permission; confirm the rights and provenance required for your intended use.
Can FlashData provide video IDs as well as files?
Yes. FlashData can supply video ID collections, dataset samples and direct dataset delivery. You can also bring your own IDs and use the APIs to operate your pipeline. Known IDs do not guarantee that every source video remains available.
How do I get a YouTube video dataset sample?
Send your requirements using the sample request on this page. Specify whether you need IDs, metadata, media or available captions. The team will confirm sample availability and contents; the layout example on this page is not an actual dataset sample.
Is everything bundled into one API request?
No. YouTube Data APIs and the Download API are separate operations. Optional download metadata is a companion JSON file; existing transcripts or captions require the relevant data API. For managed datasets, agree the requested contents with the team.