Enterprise Knowledge Search · Real-time Analytics
Refik Anadol Studio needed to process and understand billions of nature images fast enough to train a new multimodal model for DATALAND, while also enriching the dataset with better semantic context than previous open-source captioning approaches could provide.
Reported impact
Refik Anadol Studio dramatically accelerated multimodal data preparation, improved caption quality, and built a scalable data engine for an AI-native art experience that can query and reason over nature imagery at massive scale.
Refik Anadol Studio built the Large Nature Model on Google Cloud using Vertex AI, Gemini 2.5 Flash, BigQuery, and Google Cloud Storage. Vertex AI orchestrates the image-processing pipeline, handling preprocessing, object detection, and embedding generation in parallel so the team can prepare huge visual datasets far faster than before. Gemini 2.5 Flash then generates richer, more scientifically accurate captions for each image, turning raw visuals into semantically meaningful training data rather than generic tags. Google Cloud Storage holds the petabyte-scale archive and metadata record system, while BigQuery lets the team query billions of image-caption pairs in seconds for retrieval and analysis. This stack gave Refik Anadol Studio the speed, accuracy, and scale needed to turn a massive image corpus into a living, queryable nature encyclopedia that can power generative art and RAG-style interactions inside DATALAND.
Images are stored in Google Cloud Storage, then processed in Vertex AI for object detection and embeddings. Gemini 2.5 Flash creates captions for the same data, and the resulting image-caption pairs are queried through BigQuery for retrieval, analysis, and downstream model use.
Accelerated the generation of image embeddings by 7x with Vertex AI
Enabled the ability to query billions of image-caption pairs in seconds rather than days
Reduced object detection and image preprocessing time from 30 hours to just 80 minutes using Vertex AI
The combination of Vertex AI orchestration, Gemini's captioning, and the instant-access storage from Cloud Storage and BigQuery provides the technical engine
With Gemini 2.5 Flash, we improved caption accuracy by nearly 40%
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
The page's Organization JSON-LD lists numberOfEmployees value 35, and the visible page text says Refik Anadol Studio has approximately 35 employees as of August 2026.
Related
Lenovo Sample Oct 13, 2026 Formula E Formula E wanted fan experiences that feel immediate, immersive, personalized, and participatory, but its broadcast and event experience was limited by inconsistent onboard video quality, information spread across… Developed in under 30 days. AI Fan Experience Guide · AI Video Optimization · AR broadcast graphics 
Google Cloud Sample Archived · Oct 2, 2026 Grupo Paolini Grupo Paolini is an Argentine family-owned corporate group with more than 60 years of history and 950 employees across four business lines: road construction, mining services, architecture, and agribusiness. is an Argentine family-owned corporate group with 60 years of history and 950 employees across road… Gemini Enterprise · Google Workspace Google Cloud Sample Archived · Jul 31, 2026 Aible Aible needed to deploy secure enterprise AI agents quickly while keeping customer data inside each enterprise cloud boundary, avoiding idle GPU cost, and proving that large-scale analytics and model orchestration could… 1000x analytics cost efficiency gain BigQuery · Cloud Run · G4 GPUs 