Automated News Video Generation: One Edition Every Morning

A morning brief is data, not writing: today's temperature, a three-day outlook, and three headlines. Bind those 26 values to a fixed layout, render one edition on a cron before the office opens, and publish it without a review step. Each 38-second edition costs one credit.

The finished video

This guide renders News & Weather - Daily Morning Brief - a 38-second 16:9 edition in seven scenes: a "Good morning" open carrying the date, today's weather over a full-bleed backdrop, a three-day forecast row, the day's top three stories in three different layouts, and a sign-off.

Twenty-six fields change every morning. Fifty-six overlays are driven by them.

What you need

• A Renderly API key. Create one in the dashboard under Settings → API keys.
• A weather feed with current conditions and a three-day outlook.
• A headline source: your own CMS, a wire, or an RSS feed you already publish.
• A scheduler that fires once a day.

Step 1 - Decide whether your brief is data or a script

Search this and the results sell you an AI news generator: paste a story, get an avatar reading it. That is a different product, and the difference decides whether the pipeline can run unattended.

A generated brief is writing. A model drafts a script, a synthetic presenter performs it, and the output is a new video every day - which means somebody has to watch it before it goes out, because nobody ships an unreviewed machine-written script under a masthead.

A template brief is binding. The layout, the timing and the words around the data are fixed and were approved once. The only thing that changes at 6am is 26 values that came from feeds you already trust. There is nothing new to review, so the job can publish on its own, and that is the whole reason to automate it.

Pick the template route when the brief is the same shape every day. Pick a generator when each edition genuinely needs different writing - at which point you have a video production problem, not a scheduling one.

Step 2 - Pull the morning's numbers from a weather feed

Seventeen of the 26 fields are weather. They map onto a normal feed response, but three of them are formatted strings rather than raw numbers, and a feed will not guess the format:

forecast_N_high and forecast_N_low are deliberately asymmetric - the high is the big 150px numeral and the low is the small line under it, so the word "Low" lives in the value rather than in the layout. Format both in the mapping function and write one test for it; this is the field pair that ships wrong.

Step 3 - Budget the three headlines before the feed picks them

The three story scenes use three different layouts, and their headline boxes are not the same size:

The third is the trap. It is by far the widest box and by far the least forgiving, because a full-bleed layout gives the headline only two lines at 96px. The template's own three headlines run 62, 58 and 61 characters, and the boxes were drawn around that. Treat roughly 60 characters as the budget, cap it in the query that selects the stories, and put the shortest headline of the three in slot three.

An over-long headline does not fail the render. It renders, the last line falls out of the box, and the job reports success.

Step 4 - Render one edition on a morning cron

The assembled object goes out as one request. Note the template ID: news-weather-updates is the API identifier, and news-weather-daily-brief is only the gallery URL.

curl -X POST https://renderly.video/api/v1/renders \
  -H "Authorization: Bearer $RENDERLY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "templateId": "news-weather-updates",
    "replacements": {
      "edition_date": "Tuesday, August 25",
      "weather_backdrop": "https://cdn.example.com/sky/overcast.jpg",
      "current_temp": "68",
      "current_condition": "Partly Cloudy",
      "current_high_low": "H 74°  •  L 55°",
      "forecast_1_day": "Wed",
      "forecast_1_high": "76°",
      "forecast_1_low": "Low 57°",
      "forecast_1_cond": "Sunny",
      "news_1_category": "World",
      "news_1_headline": "City leaders approve sweeping transit plan to cut commute times",
      "news_1_image": "https://cdn.example.com/stories/transit.jpg"
    },
    "webhookUrl": "https://your-app.com/hooks/renderly"
  }'

The job around it is a mapping function and a schedule:

// 05:30 local, once a day.
const [weather, stories] = await Promise.all([
  weatherFeed.today(),
  cms.topStories({ limit: 3, maxHeadlineLength: 60 }),
]);
 
await fetch("https://renderly.video/api/v1/renders", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.RENDERLY_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    templateId: "news-weather-updates",
    replacements: {
      edition_date: formatEditionDate(new Date()),
      ...toWeatherFields(weather),
      ...toStoryFields(stories),
    },
    webhookUrl: "https://your-app.com/hooks/renderly",
  }),
});

maxHeadlineLength in the query is the whole of step 3, enforced at the only place it can be enforced cheaply. Any field left out of replacements keeps the template's default, so a feed that returns nothing publishes yesterday's sample weather rather than failing.

Step 5 - Publish the edition before the news goes stale

Because webhookUrl was included, Renderly posts back when the file is ready:

{
  "event": "render.completed",
  "data": {
    "jobId": "cm9x2k4p10001qw8r7a3b2c1d",
    "status": "COMPLETED",
    "outputUrl": "https://cdn.renderly.video/renders/cm9x2k4p10001qw8r7a3b2c1d.mp4",
    "creditsUsed": 1,
    "durationInFrames": 1140,
    "width": 1920,
    "height": 1080,
    "fps": 30
  },
  "timestamp": "2026-08-25T05:31:07.902Z"
}

Push outputUrl straight to the site, the newsletter or the channel. There is no approval queue in this pipeline by design: the layout was signed off once, and the only new material is the data. Payloads are signed with HMAC-SHA256, so verify the signature before you trust the body - the webhooks guide covers the check.

What this costs at scale

Renderly bills 1 credit per minute of 1080p output, rounded up to the nearest half credit. A 38-second edition rounds to one credit.

Ten markets, every morning, all year, fits inside the $29 plan at roughly ten cents an edition. The interesting number is the last row: a fifty-market daily video brief costs $99 a month to render, which is less than one freelance edit, and the marginal cost of the fifty-first market is a row in a config file.

Where to go next

The pattern here is a schedule with a feed behind it. The TikTok guide runs the same cron shape against a booking system rather than a weather API, and the Google Sheets guide covers the case where a person, not a feed, fills the row.

If step 1 is still the open question, the long version of that argument - with the failure modes of each approach - is in template-based video vs AI-generated video.