The Publisher is a digital-first news and media organization with an established web and app audience, competing for attention in a market where short-form video is now the dominant way readers consume news. Its existing content and distribution stack, CMS, publishing workflow, and app/web delivery, was already built and hosted on AWS, and was not something the Publisher wanted to replace or migrate. However, they faced significant hurdles:
- Studio-Capped Production: Anchor-led video coverage required a studio setup, an on-camera presenter, and a production/editing cycle that could not scale to cover the full volume of stories the newsroom's editorial team was already reporting on in text. Video was reserved for a small number of major stories chosen manually each day.
- Credibility & Trust Risk: Automated news content carries real credibility risk if handled carelessly: a video that misstates a fact, or can't show where its information came from, does real damage to a news brand's trust with its audience. Any automation had to make content faster to produce, not less trustworthy.
- Need for Both Automation and Editorial Control: Routine, data-driven stories needed full automation to hit scale, but higher-sensitivity stories still required a human hand on script, framing, and selects before publishing.
- No Disruption to a Live Media Stack: The brief required all of this to plug into the Publisher's already-live, AWS-hosted media application and CMS, without a disruptive migration of infrastructure readers were actively using.
Varcio designed and built the platform end-to-end, product design, backend and frontend development, and the underlying Google Cloud infrastructure, and integrated it directly into the Publisher's existing AWS-hosted media application, so publishing a generated video into the Publisher's live app required no change to how readers already consumed content:
- Continuous Ingestion & Fact Aggregation: News and event data ingested continuously from external feeds and APIs, wire services, structured event data, and other source material the newsroom already relied on.
- Gemini-Powered Script Generation: An AI-driven script and narrative generation step, built on Google's Gemini models, turned source material into a structured, fact-grounded video script, with every claim linked back to the source data it came from rather than generated freeform.
- AI Anchor Video Generation Engine: Rendered the script into a presented video news segment with a consistent virtual anchor, local language support, regional accent variation, and multiple audio/voice variations, so the same story could be delivered in the tone and language appropriate to different audience segments.
- Auto / Customize Editor Console: Editorial staff could let a story run fully automatically end-to-end, or step in and customize script, framing, or selects before a video was generated and published, the platform was explicitly built to support both modes, not to force full automation.
- Fact & Authenticity Verification Layer: Kept every generated segment traceable to its underlying source data, so the newsroom could show, on request, exactly what a claim in a video was based on, directly addressing the credibility risk that concerned the Publisher going in.
- Analytics & Generation-Provenance Dashboard: Gave editorial and product teams visibility into what was generated, how, from what sources, and how each video performed with the audience after publishing.
- AWS Publishing Integration: Finished videos delivered directly into the Publisher's existing AWS-hosted media application (EC2) and CDN/publishing layer (CloudFront + MediaLive), sitting alongside the existing stack rather than replacing it.
Fact Generation Accuracy & Automated Clip Approval
Mapping the newsroom's existing story pipeline, source feeds, and AWS-hosted publishing stack to design a Google Cloud generation pipeline that could plug in without disruption.
Building the Gemini-based script and narrative generation step, with every claim linked back to its source data rather than generated freeform.
Building the AI anchor video generation engine, with multi-language and multi-accent support, alongside the fact and authenticity verification layer that keeps every segment traceable to source.
Building the auto/customize editor console and analytics dashboard, integrating publishing directly into the Publisher's live AWS media application, and rolling out to the newsroom.
The newsroom gained the ability to produce video coverage for a much wider range of stories than the studio-based workflow could ever support, without a proportional increase in production staff. Seventy percent of automatically generated clips were approved for posting as-is, meaning the majority of routine, data-driven stories required no editorial rework before publishing, editorial time was concentrated on the remaining 30% and on higher-sensitivity stories by choice.
Local language support, regional accent variation, and multiple audio/voice options allowed the same underlying story to be delivered appropriately across the Publisher's different audience segments, rather than a single generic voice and language. The fact and authenticity verification layer, combined with the 90% generation accuracy achieved, gave the Publisher's editorial leadership confidence to expand automated coverage, because every generated segment remained traceable to its source data rather than being a black box. The platform was delivered in approximately 4 months by an 8-person cross-functional Varcio team, and added onto the Publisher's existing AWS-hosted stack rather than requiring a disruptive migration of an already-live media application.
"70% of automatically generated clips were approved for posting without any editorial rework, at 90% underlying generation accuracy."