Smarter Metadata Starts with Smarter Models: What's New in Fingerprint+ v13.4

We recently released Fingerprint+ v13.4 – the latest version of the AI model that powers keyword, mood tag, and genre classification across the Vionlabs platform. Here's what changed and why it matters.

Why Model Updates Matter for Metadata Quality

Fingerprint+ is the core model behind Vionlabs' metadata enrichment. It's what turns raw video content into structured, searchable intelligence – the keywords, mood tags, and genre signals that feed into recommendations, editorial curation, and content discovery.

When this model improves, everything downstream improves with it. More accurate keywords mean better search results. More precise mood tags mean more relevant recommendations. And fewer errors mean editorial teams spend less time correcting AI output and more time doing creative work.

With v13.4, the improvements focus on two areas: reduced hallucination and better understanding of niche content.

Improved precision

The previous version of Fingerprint+ had a tendency to default to familiar, "safe" tags. Version 13.4 is significantly more precise

The Terminal

A well-known comedy-drama about a man stranded in an airport – warm, quirky, and human.

Version 13.4 now captures more nuanced tone of the film: immigration, friendship, identity, and resilience as keywords, with mood tags like thought-provoking, heartwarming, humorous, and uplifting.

When metadata is wrong, content gets surfaced in the wrong context. A viewer looking for a feel-good comedy shouldn't land on a survival horror film – and vice versa. Across a catalogue of thousands of titles, these corrections compound into meaningful improvements in recommendation quality and discovery accuracy.

Better Understanding of Niche Content

Not all content fits neatly into a single genre box. Some of the most interesting titles – and some of the hardest to classify – sit at the intersection of multiple genres and tones. This is where AI models tend to struggle the most, defaulting to the dominant genre and ignoring the rest.

Fingerprint+ v13.4 handles these hybrid categories with more nuance.

Violent Night

This is a Christmas action-comedy with strong horror elements – holiday cheer meets brutal fight sequences.

Version 13.4 captures tone complexity. Keywords like christmas, holiday, santa claus, and family now sit alongside villains, hostages, action fights, and home invasion. Mood tags show the same duality: festive and light-hearted next to dark humor, satirical, and thrilling.

This matters because niche content often has the most engaged audiences. Fans of holiday horror or action-comedies know exactly what they're looking for. If your metadata can match that specificity, you surface the right content to the right viewers – rather than burying it in a generic "Comedy" or "Action" rail where it disappears.

Beyond genre-blending films, v13.4 also brings improved differentiation across specific content domains. Sports content, for example, now benefits from better classification – the model can more reliably distinguish between basketball and baseball footage, which matters for platforms managing large sports libraries or highlight catalogues.

How We Validate Quality

One question we get asked often is how we measure the quality of a model update like this. Currently, every Fingerprint+ release is validated by our in-house content experts – people who understand genre, tone, and storytelling at a deep level. They review outputs across a wide range of titles and content types to ensure the metadata meets a human standard before it reaches customer catalogues.

We strongly believe that expert human review should always be part of the process. AI-generated metadata needs to make sense to the people who actually curate and programme content, not just to a statistical score.

What This Means for Customers

Fingerprint+ v13.4 is available now. Customers can adopt the new model version on their own schedule, with support from the Vionlabs team to ensure a smooth transition. The updated model is fully integrated with the Vionlabs portal, so once you're on v13.4, the improved metadata flows directly into Editorial Lab and your API outputs.

These kinds of updates don't come with a splashy product announcement. But they represent something fundamental to what makes Vionlabs different: we build and continuously improve our own AI models, trained specifically on entertainment content.

Generic AI models treat a horror film and a romantic comedy as roughly the same classification problem. Domain-specific models understand that the difference between eerie and heartwarming isn't just a label – it's the difference between a viewer clicking play or scrolling past.