HP came to us needing 2,500 product videos across 28 markets. The agency quote was $1.25M and 6 months. We said 7 minutes and $50 per video. They said yes.
That's the story of Velocity.
The unit economics of e-commerce video hadn't moved in a decade. Studio production ran $500–$2,000 per video with a 2–3 week turnaround per SKU. For a catalog spanning thousands of products across 28 markets, the math to keep everything current didn't close. Brands were choosing between coverage and cost.
The problem ran deeper than catalog videos. Marketing and ad teams needed dozens of content variations for every campaign — different formats, copy treatments, and crops — at a speed their paid channels actually demanded. With the traditional process taking weeks per asset, running ads at any real velocity was impossible.
The brief was specific: broadcast-quality output at the cost and speed of image generation. No studio, no actors, no 6-week calendar block. The quality bar wasn't the problem — the pipeline that produced it was.
Three things had to be true simultaneously: quality had to be indistinguishable from studio output, speed had to be under 10 minutes per video, and the pipeline had to support 28 languages without re-recording.
Existing tools handled one of these. None handled all three.
We also had to make it accessible to non-technical brand managers rather than engineers, because the people generating videos at HP were marketers, not ML teams.
A user pastes a product URL or uploads texts and images — including mobile shots. The pipeline segments each image automatically, stripping backgrounds to produce clean silo assets. Lifestyle compositions are then generated from the same uploaded visuals.
Simultaneously, the system extracts primary and secondary accent colours directly from the product images, and parses the brand kit to capture visual identity and tone of voice.
Product category, extracted text, and processed images are scored against a library of thousands of curated templates using a weighted matching algorithm — balancing category fit, visual style, layout, and brand consistency. The top-scoring template wins.
Text and images are swapped in, AI voiceover is generated, and the video renders. From URL to finished asset: under 7 minutes.
Velocity is an AI video generation pipeline built on top of Avataar's 3D rendering engine. Input: a product URL, images, and a brand kit. Output: broadcast-quality videos and ready-to-run ads in under 7 minutes.
Brands could generate dozens of content variations rapidly — different formats, copy treatments, and visual styles — at the speed their paid channels required. A native Meta integration let teams push finished ads directly from the platform to ad manager, without re-uploading or leaving the tool.
Performance data fed back into a recommendation engine that surfaced templates and content combinations that had worked best for similar merchants. The platform got smarter with every campaign that ran through it.
A major enterprise client used it to generate their entire product catalog across 28 markets in weeks rather than years.
Enterprise contract signed for 2,500 videos across 28 markets. Full catalog delivered in weeks — versus the 18+ months the agency model would have required at that volume.
25,000 users signed up in 6 months after public launch, with no paid marketing. All organic, mostly word-of-mouth from the enterprise launch.
Video conversion rates ran 3–5x higher than static images. That performance data became the sales artifact that made the next enterprise conversations easy. Cost per video: $50 versus $500–$2,000 at agencies — a 10–40x reduction that changed the ROI framing entirely.
We built Velocity for one enterprise client first and the market second. The pipeline was heavily optimized for their SKU structure and ERP data format. When we tried to onboard the next enterprise client, we had to rebuild 40% of the ingestion layer. I'd have designed for a configurable data schema from day one, even if it meant shipping two weeks later.
The other miss: we never built a self-serve tier. 25k signups sat in a waitlist because onboarding was fully manual. That was a missed growth unlock. We got the enterprise contract. We left the bottom of the funnel on the table.
The best product insight isn't always technical. The real unlock for Velocity wasn't a better model. It was realizing that brand managers needed a tool that felt like Canva, not like Blender.
Speed and quality were table stakes. Simplicity was the moat. If the person generating the video has to open a terminal, you've already lost.
Video at scale isn't a creative problem. It's a pipeline problem. Once you treat it like infrastructure — inputs, transforms, outputs, quality gates — the cost and time numbers become inevitable.
The creativity happens at the edges. The middle is just engineering.
Building AI products since 2020, from spatial computing to finance automation.
