Executive Summary
from idea to working pipeline
AI-orchestrated
text-to-video, interchangeable
from date to finished video
The Project
DailyMix produces short educational videos for digital signage: every day, screens in public spaces show a clip about an event in U.S. history that took place on that date. Made by hand, such content requires research, a script, footage, captions and editing, every day of the year.
The goal was to automate the whole chain, from a calendar date to a finished, captioned video ready for broadcast, leaving the editor to review and fine-tune the final cut.
What Was Built
Event selection and script — a language model chooses the most significant event in U.S. history for the given date, establishes the year and writes a narrative divided into segments, with timed captions that together answer three questions: what happened, why it matters and what followed.
- Calendar logic for U.S. holidays, both fixed and movable, so that key days get the right subject
- Prompt refinement — the language model improves the visual brief for each segment in several passes: period, setting, props, action, camera, lighting and style
- Three interchangeable video models — OpenAI Sora 2, Kling and Google Veo behind a single interface, so the best model can be chosen for each subject and the pipeline does not depend on one vendor
- Captions generated in the standard SRT format and overlaid on the video
- Assembly of the segments into one clip with smooth transitions
The Method
One engineer working with an AI assistant. The pipeline was designed as a sequence of independent stages that exchange data through files, so that each stage (script, prompts, video, captions, assembly) can be run, inspected and improved on its own. This made it practical to compare video models on the same script and to tune the prompts stage by stage.
From idea to a working pipeline that produces finished videos took one month.
Tech Stack
Conclusions
Video generation models are advancing quickly, but the value lies less in any single model than in the chain around it: choosing the subject, writing a script that is historically sound, turning it into precise visual briefs and assembling a clean result.
One engineer. One month. A pipeline that turns a calendar date into a finished, captioned educational video, with three video models to choose from.