AI video ads: script first, footage second, scene by scene approval
AI video ads are built by writing and approving a script and scene text before any footage is produced. Scenes are built from ordinary work that happens inside most companies, which is where the realism comes from. Footage and motion are produced with AI tools, then editing, music, captions and on screen text are added. A voiceover is optional, and the advert is cut into separate formats per channel.
A video advert usually starts with footage. Something is filmed or generated first, and the words get added to it afterwards. AI video ads reverse that order. The work starts with a script and scene text. It sets out what each scene shows and what order the scenes run in. It also fixes what line appears on screen, worked out on paper first. That script goes to the client and no footage is produced before it is approved. When footage comes first, the words get forced to fit the picture. When the script comes first, the footage stays in service of it.
In an advert aimed at a general audience, realism does not come from an invented scene. It comes from ordinary work that happens in most companies. Scenes are built from ordinary moments found in most companies: a phone ringing, a meeting starting, a door opening. These are moments most viewers already recognise. That choice is deliberate. A viewer who recognises a scene finds it easier to connect the advert to their own experience. While the script is written, what each scene is there to say gets fixed. The plan goes into the approval round in that shape.
Once the script is approved, footage and motion are produced with AI video generation tools. The raw footage produced at this stage is not an advert on its own. Editing, music, captions and on screen text are added on top, and these layers turn a scene into something watchable. A voiceover is an optional part of that final mix. A business can carry its message through on screen text alone, with no voiceover at all. Which scenes carry a voiceover and which carry only on screen text is decided together at the script stage, not during production.
The video does not end in a single format. The same script is cut separately per channel into vertical, square and horizontal formats, covering platforms such as Reels, Shorts and TikTok. Each cut is edited to fit the frame and viewing habit of its platform. A real person's face or voice is used only with that person's written permission. AI generated content that goes live is labelled according to the relevant platform's rules. Every scene goes back to approval once it is produced. If a fix is needed, that fix covers only the scene in question, not the whole video. Once the video is set to run as a paid advert, that step connects to the ad management service. ALTAI Digital works from Rotterdam and Istanbul.
What goes wrong today
- 01
Footage gets produced first, words are added after
When footage is filmed or generated first and words get added afterwards, the line ends up forced onto the scene. In AI video ads the order runs the other way: script and scene text are written and approved first, footage comes only after that. That order keeps the footage in service of the words.
- 02
A scene stays invented and a viewer cannot connect to it
An abstract scene, or one that matches no business in particular, feels distant to a viewer. Scenes in an advert aimed at a general audience are built from ordinary work inside most companies. That is where the realism comes from. The script fixes which everyday moment becomes which scene.
- 03
A scene enters production without approval
If footage is produced before the script is approved, the client only sees the video once it is finished. Any change then costs more. In AI video ads every scene goes to approval. A fix runs at the scene level, not the whole video.
- 04
A single format gets reused on every channel
A video cut only horizontally or only vertically and dropped onto every channel ends up cropped or sitting in empty space. Formats are cut separately as vertical, square and horizontal, so each channel gets its own cut from the same script.
- 05
A real person's face or voice is used without consent
Using an employee's or a third party's image or voice in a video without their consent harms both that person and the brand. In AI video ads a real person's face or voice only appears with their written consent.
- 06
AI generated content goes live without a label
Footage produced with AI tools that goes live without the label the relevant platform requires can get the account flagged. AI video ads follow the AI content labelling rules of the relevant platform, and that check runs before anything goes live.
Scope of the build
Script and scene text
The work starts with a script and scene text, worked out on paper first. It sets out what each scene shows, the order scenes run in and what line appears on screen. This text goes to the client before any footage is produced.
Scenes built from everyday moments
Scenes in an advert aimed at a general audience are built from ordinary work found in most companies. A moment most viewers recognise connects the advert to the viewer. The script fixes which everyday moment becomes which scene before production starts.
Script approval
The script and scene text go to the client, and no footage is produced before it is approved. At this stage approval covers the whole video; footage production only starts once it is signed off.
Footage and motion production
Every scene in the approved script has its footage and motion produced with AI video generation tools. The raw footage produced here is not an advert on its own; it is completed in the following step.
Editing, music, captions, on screen text
Editing, music, captions and on screen text are added on top of the produced footage, turning a scene into something watchable. A voiceover is optional, and the message can be carried through on screen text alone.
Scene approval and fixes
Every scene goes back to approval once it is produced. If a fix is needed, that fix covers only the scene in question, not the whole video from the start.
Format variety and channel cuts
The same script is cut separately per channel into vertical (Reels, Shorts, TikTok), square and horizontal formats. Each cut is edited to fit its own platform's frame.
How the build runs
- 01
Script and scene text writing
What each scene shows and what line appears on screen is decided, and scenes are built from everyday work moments.
- 02
Script approval
The script and scene text go to the client, and no footage is produced before approval.
- 03
Footage production and completion
Footage and motion are produced with AI tools, then editing, music, captions and on screen text are added.
- 04
Scene approval and channel cuts
Every scene goes to approval, fixes run at the scene level, and the same script is cut separately per channel.
Products used in this build
Terms used on this page
- Scene text
- Scene text is the written plan that sets out what a scene shows and what line appears on screen before footage is produced.
- Script approval
- Script approval is the stage where the scene text goes to the client and is signed off before footage production starts.
- Scene approval
- Scene approval is the step where each produced scene is sent back to the client for sign off after editing and additions are complete.
- Channel cut
- A channel cut is a platform specific version of the same script, prepared separately for vertical, square and horizontal formats.
- On screen text
- On screen text is the written line added over the video that carries the message even without a voiceover.
AI video adsFrequently asked questions
AI video ads are built by writing and approving a script and scene text before any footage is produced. Footage is made afterwards using AI video generation tools. Editing, music, captions and on screen text are added on top of it.
If footage is produced first and words are added afterwards, the line ends up forced onto the scene. Writing and approving the script first keeps footage production in service of the words. This order sets out what each scene says before anything is produced.
Scenes in an advert aimed at a general audience are built from ordinary work inside most companies, such as a phone ringing or a meeting starting. The realism comes from these scenes, since a viewer recognises a moment they already know.
No. A voiceover is optional, and the script can be carried through on screen text alone.
The same script is cut separately per channel into vertical (Reels, Shorts, TikTok), square and horizontal formats. Each cut is edited to fit the frame and viewing habit of its own platform.
Only with that person's written consent. This rule applies the same way to every scene; without consent, that person's image or voice does not go into the video.
Every scene goes back to approval once it is produced. If a fix is needed, that fix covers only the scene in question, not the whole video from the start.
Once the video is set to run as a paid advert, that step connects to the ad management service. Before it goes live, the platform's AI content labelling rule is checked for compliance.
The process first, the proposal after
In a short discovery call we map the process together and tell you plainly whether it is worth automating. If it is not, we say that too.
30 minutes · no preparation needed
