Worked Case: Anatomy of a 30-Second AI-Assisted Explainer

This case follows one finished piece from blank page to upload. The issue is invented: the Cold Springs Water District has proposed raising the base residential rate from $42 to $58 a month, starting July 1, to replace a 1962 main line. The hearing is October 28. A residents' group wanted a neutral 30-second vertical explainer telling neighbors what is proposed and how to weigh in. Every stage below shows where AI did the work and where it was kept out.

Stage 1: Facts before pixels

The group listed every fact the video would state and where it came from. In real life, that is the district's rate notice and board packet.

Stage 2: A 75-word script

Narration at a calm 150 words per minute is 2.5 words per second. Thirty seconds holds 75 words. A beat at the start and end leaves room for 68 to 72.

Cold Springs, your water bill may go up.
The district wants to raise the base rate
from forty-two dollars to fifty-eight dollars a month.
That's thirty-eight percent, or about
a hundred ninety-two dollars a year.
The money would replace a main line
laid in nineteen sixty-two.
The board holds a public hearing
October twenty-eighth at six p.m.
Written comments are due the twenty-seventh.
Details and the full notice are linked below.

Stage 3: Shot list

Nine shots, vertical 9:16, about 30 seconds, mixing AI stills, a little AI motion and material the group controlled.

Note shot 7. The district office is a real place, so it gets a real photo. AI handles the generic and the symbolic. Reality handles the real.

Stage 4: Images

The style sheet came first (lesson 3). Shots 1, 2, 5 and 8 were generated in ImageStudio at 1080x1920 on Draft, with Final used only for the keepers.

STYLE: flat vector illustration, clean shapes, light paper grain
PALETTE: slate blue, ochre, off-white, charcoal
LIGHT: soft morning light, gentle shadows
CHARACTER: Rosa, woman about 60, short gray hair, round wire
  glasses, olive green canvas jacket, red scarf
FRAME: tall vertical, subject in middle third, open space at
  top for titles, plain lower fifth
NO: text, logos, real place names

Shot 2's prompt was "Rosa sitting at a small kitchen table opening a white envelope, coffee mug beside her, window behind her" plus the style sheet. With "Show platform safe areas" on, any image with Rosa's face in the bottom fifth was rejected.

Stage 5: Voice and music

The script went into Text2Speech in Voice Studio, where two or three stock voices were compared and a calm, neutral one chosen. "Nineteen sixty-two" and "Cold Springs" are the words most likely to come out wrong, so those lines got a careful listen and a regeneration where needed. The target length was 27 to 29 seconds. When a take ran long, words were cut rather than speeding the voice past natural.

In MusicStudio, the "Instrumental (no vocals)" mode and a style like "instrumental, soft piano and light acoustic guitar, 90 BPM, steady, calm, no big builds" produced two songs, and the steadier one won. The license terms for the Mureka plan decide whether it can be used in anything paid (lesson 8).

Stage 6: Assembly

The pieces came together in AVPrep3, the desktop timeline editor, in four lanes:

• Lane 1: images and the real photo, in shot order, as plain holds. ("Images to Video" turns stills into clips; the AVPrep3 help does not describe an animated zoom.)
• Lane 2: titles from "Add Title": $42, $58, 38%, $192 a year, Installed 1962, Hearing Oct. 28, 6 p.m. Each was checked against the fact sheet.
• Lane 3: narration, with each title lined up to the word it matches.
• Lane 4: music, ducked under speech, faded in and out.

Captions were transcribed from the narration, corrected against the script and burned in above the bottom app zone; the AV chapter text people can read covers size and contrast. Straight cuts joined the shots, since crossfades between AI stills look like melting.

VideoStudio offers a simpler route: "Create Video" builds a slideshow from up to 10 images, and "Replace video audio" lays in a narration-plus-music mix made in SoundStudio. Titles and captions are harder there, so they would be baked into the still images in an image editor first.

Stage 7: Labels

• On screen, whole video: "AI illustration" small in the top corner, inside the safe area, about 28 to 32 pixels tall on a 1080-wide frame.
• End card: "Illustrations: AI-generated (ImageStudio). Narration: AI voice (ElevenLabs). Music: AI-generated (Mureka). Photo of district office: [member's name]. Source: Cold Springs Water District rate notice."
• Post text: the same sentence plus the link to the notice.
• Platform labels: the illustrations are stylized, not realistic, and Rosa is invented. YouTube's policy does not require its AI label for clearly unrealistic content. Its rule names AI music that is the main focus of a video, but its examples list "AI generated music" with no such qualifier, so a background bed is a gray area. The group disclosed anyway. On TikTok, a label sticker or caption costs nothing.

Stage 8: Export and log

The export was a 1080x1920 MP4 from an AVPrep3 platform preset. It got two viewings on a phone: once with sound off (do captions carry it?) and once with sound on (are the numbers clear over the music?). The AV chapter getting it out the door covers delivery checks.