Reinventing Creative Production with AI: How Shining Builds Fast, Large-Scale Video Search with Zilliz Cloud
<300 ms Latency at Scale
Delivers near-instant retrieval even with a library approaching 10 million vectors
>100 million vectors
Scales seamlessly to future growth with no re-architecture, no downtime, and no additional operational burden
65% less communication time
Accelerates large campaign production by cutting proposal timelines from 7 days to just 2
40% higher approval rates
Helps agencies win more pitches by surfacing highly relevant clips in minutes
We tested every mainstream vector database, and Milvus delivered the best overall performance.
Su Wei
About Shining
Shining is an AI-powered creative platform used by directors, producers, and agencies to speed up the process of turning early ideas into full client proposals. Instead of manually searching through folders or stock sites, creatives simply describe the scene they want in natural language—such as “a slow-motion close-up of a runner at sunrise”—and Shining instantly returns the most relevant video references. From there, users can turn references into storyboards within minutes and generate presentation-ready decks in the same workspace. Today, more than 2,000 creative teams now use Shining to cut hours (or even days) out of their pitch development process.
To make this possible at scale, Shining integrated Zilliz Cloud, the fully managed vector database built on Milvus. Before Zilliz, Shining’s team struggled with the same challenges many media and content companies face: huge video libraries, inconsistent tagging, slow search, and heavy engineering time spent maintaining infrastructure. With Zilliz Cloud, Shining built a fast and highly accurate semantic search engine that handles tens of millions of video assets and returns results in milliseconds. Creatives can now find the exact shot they imagine—whether defined by mood, camera angle, lighting, or style—without relying on manual metadata.
What Exactly Do Creative Teams Need from AI
Creative advertising remains one of the most manual and time-consuming parts of the media industry. Something as routine as preparing a TV commercial concept can take days—most of it spent searching for usable reference footage. Online platforms often mix high-quality clips with irrelevant material, and even when the right content exists, teams still spend hours digging through folders and scrubbing long timelines to find the exact shot they need.
The bottlenecks don’t end there. Once references are collected, teams still rely on hand-drawn storyboards, a process that’s slow, repetitive, and prone to inconsistencies. The proposal stage adds even more friction: formatting slides, managing revisions, collecting feedback, and keeping everyone aligned often leads to long nights and multiple rounds of rework. These operational hurdles drain valuable time that could be spent on actual creative thinking.
Shining recognized that to truly modernize this workflow, creative teams need an AI system built around three core capabilities:
Million-Scale Intelligent Video Search — Users should be able to type a visual description or upload an image and instantly search across millions of videos, including global ads, film storyboards, and TVCs. Automatic similarity search based on composition, color, and style can boost inspiration efficiency by as much as 80%.
One-Click AI Storyboard Generation — From a script or reference clip, the AI should generate storyboard sketches and shot-level notes—framing, camera movement, and timing—with export options ready for client use. Shining is also building a Storyboard Agent that helps refine pacing and narrative automatically.
10-Minute Proposal-Ready Slides — Storyboards, references, and strategy should come together into a polished pitch deck in minutes, with built-in collaboration and version control so teams spend less time formatting and more time creating.
How Shining Shapes Its AI Workflow
With the core challenges and needs identified, Shining refined its AI product around three practical principles that any creative team can relate to.
Designing for Real Creative Roles
Shining builds features based on what different users actually need—directors, designers, producers, and brand marketers. For example, directors get precise storyboards and high-quality visual references to support their vision. Brand and strategy teams get faster, more reliable proposal creation so they can respond to clients with confidence. Each feature is designed to reduce friction for a specific role, not just add more tools.
Supporting the Entire Production Workflow
Instead of solving just one part of the creative process, Shining focuses on the three biggest bottlenecks: finding references, generating storyboards, and assembling proposals. Intelligent search, automated creation, and real-time collaboration all work together in one integrated workflow. The upcoming Storyboard Agent will push this further, helping teams move from a brief to a polished output without jumping across multiple apps.
Defining Human–AI Collaboration Clearly
Shining also sets clear boundaries for how humans and AI work together. The AI takes over repetitive, time-consuming tasks—like searching for materials, organizing structure, or generating draft storyboards—so teams can focus on creative direction, strategic thinking, and client communication. This balance ensures speed without losing the human judgment that great creative work depends on.
The Solution: Powering Creative-Grade AI with Zilliz
Shining’s technical strategy starts with a simple idea: rebuild the creative workflow using Retrieval-Augmented Generation (RAG). First, help users instantly find the right visual references; then use those references to generate high-quality storyboards and proposals. To do this reliably, Shining needed vector search that could grow from tens of millions to hundreds of millions of video embeddings without slowing down or becoming too expensive to operate.
In the early stages, Shining built its system on open-source Milvus, confirming that vector search was the right foundation for creative-reference use cases.
“We tested every mainstream vector database, and Milvus delivered the best overall performance,” says Su Wei, CTO of Shining.
But as Shining’s library approached 10 million vectors, maintaining its own cluster became increasingly difficult. Memory usage spiked, operational overhead grew, and engineering time shifted away from product development toward keeping infrastructure stable. To break through this limit—and prepare for much larger datasets—Shining migrated to Zilliz Cloud, the fully managed Milvus service. The shift removed all infrastructure burdens while giving the team the elasticity and reliability needed for long-term scale.
With Zilliz Cloud, Shining now benefits from three key capabilities:
High-Performance Vector Search: Even with nearly 10 million vectors, Zilliz Cloud keeps search queries under 300 ms, giving users a near-instantaneous retrieval experience.
Elastic Scalability: Shining expects its library to grow to hundreds of millions of vectors by 2025. Zilliz Cloud’s architecture is built to scale to this size smoothly—with no re-architecture, no downtime, and no operational surprises.
Tiered Storage for Cost Efficiency: Zilliz Cloud introduces tiered storage, where hot data stays in memory for fast access, while less frequently used assets are automatically stored in S3. This allows Shining to maintain strong performance while controlling storage costs as the dataset expands.
The Results: Real Impact for Creative Teams
Since launch, Shining has powered the workflows of more than 2,000 creative teams, delivering measurable gains across real production environments.
Faster production for large campaigns: In an automotive TVC project, Shining helped a director’s team cut the proposal timeline from 7 days to just 2—while reducing communication time by 65%.
Higher win rates for agencies: For 4A teams working on BYD’s CarShow project, Shining surfaced more than 300 highly relevant reference clips in minutes, contributing to a 40% increase in proposal approval rates.
Bigger output for small studios: For 10-person creative team, the platform has become a force multiplier—boosting annual proposal output from 50 to 200 decks and enabling a 3× increase in team productivity.
Across agencies, studios, and production teams, Shining is now a trusted technology partner, providing standardized, AI-powered tools that streamline the entire creative workflow—from the first idea to the final client deck.
Conclusion
Reflecting on Shining’s growth, CTO Su Wei attributes their success to a combination of timing, technology, and focus. Large models finally made it possible for AI to understand creative assets in a way manual tagging never could. At the same time, modern vector databases made large-scale semantic search practical—providing the technical foundation Shining needed to rethink how creative teams work.
Choosing Zilliz Cloud as its vector database was a pivotal decision. The platform’s performance, reliability, and built-in scalability aligned directly with Shining’s long-term needs, allowing the team to grow from millions to tens of millions of vectors without being slowed down by cluster tuning or infrastructure management. With operations handled by Zilliz Cloud, Shining’s engineers could stay focused on product innovation instead of backend maintenance.
Equally important, Shining stayed laser-focused on the people who create the work—brand teams, agencies, directors, and designers. By clearly defining what AI should automate and what humans should lead, Shining built tools that genuinely reduce friction rather than add complexity. This combination of the right technology and a deep understanding of user workflows continues to give Shining an edge.
With this foundation in place, Shining is moving faster than ever—building AI tools that meaningfully enhance creative work and helping teams across the industry produce better ideas, in less time, with less effort.