Automated vehicle video generation

A branded video for every vehicle

Video becomes one more automatic output, not a separate production job.

What we did CodeLeap built a rendering engine on Remotion where each video is composed in React code from a parameterised template, a render API the client's platform calls per vehicle, and serverless rendering on AWS Lambda that brought the cost per video down to where a video for every listing made commercial sense.

Automated vehicle video generation — Automation case study

Every listing becomes a video.

Our client works in automotive imaging and wanted a branded promotional video for every vehicle listing, at a volume no editing team could match. CodeLeap built the video system end to end: a rendering engine on Remotion, where every video is composed in React code from a parameterised template, a render API their platform calls for each vehicle, and the AWS infrastructure to produce them at scale.

Product features

Videos composed in code

Every video is assembled programmatically with Remotion, using image, sequence, map and video-in-video slides, a library of label styles and transitions, and zoom behaviour tuned per shot type. Consistency is maintained across every output, and adjustments are made in code rather than in an editing timeline.
A render API for the platform

The client's platform requests videos through an API, passing parameters such as slide types, text, colours, fonts, aspect ratio and style presets, and receives finished MP4 files. CodeLeap provided the documentation their development team integrated against directly.
Computer-vision-driven effects

The platform's object detection identifies features such as headlights, wheels and seats, which drives animated highlights and zooms that move between real parts of the vehicle. Number plates are blurred or substituted within video frames using keypoint detection validation.
Serverless rendering at scale

The rendering engine was benchmarked extensively. By 2021 a 24-slide video rendered in roughly 2.5 minutes on a single EC2 instance, producing around 500 videos a day, and the system was later rebuilt on AWS Lambda for serverless rendering, with S3 managing assets and DynamoDB tracking every job.

Challenge

Promotional video sells vehicles, but producing it by hand is only viable for a handful of listings. The client needed a video for every vehicle they processed, which meant the whole production had to happen without an editor touching it: choosing which shots to use, sequencing them, adding branding and specifications, animating the highlights, and rendering the file.

Composition was only half the problem. Rendering video is expensive, and a cost per video that looks reasonable in testing becomes prohibitive at production volume, so the system had to be measured and re-engineered around throughput and cost rather than simply made to work.

Solution

CodeLeap built the video generation system end to end. Videos are composed in React code with Remotion, so a template defines the structure and each vehicle's video is produced by passing its own data through it. The client's platform calls a render API for each vehicle and receives a finished MP4, with slide types, text, colours, fonts, aspect ratio and style presets all controlled per request. Object detection from the client's platform drives animated highlights that move between real vehicle features, and number plates are handled automatically within the frames.

The rendering infrastructure was benchmarked and then rebuilt on AWS Lambda for serverless rendering, with S3 for assets and DynamoDB for job tracking, which took the cost and throughput to where producing a video for every listing made commercial sense. The system went live in production and has run since, generating listing videos with no human editing step.