fairytale video
Make an ad
How to make an AI ad: a step-by-step guide
An AI ad is a video ad in which the person, the setting and the motion are generated instead of filmed. The generating takes minutes. What decides whether the ad is any good is the hour of decisions before it: what the ad is for, what's true about the product, what gets said and what happens in each second.
This guide goes through those decisions in order, from the idea to the checks you run before the ad goes out. It assumes a short vertical video, 10 seconds or less, for TikTok, Instagram Reels or YouTube Shorts.
1. Decide what the ad has to do
Write one sentence: who the ad is for, what they should take away and what they should do next. "Show runners that the bottle fits in a jacket pocket" is an ad. "Introduce our brand" isn't. Ten seconds has room for one idea, so pick the strongest one and keep the rest for the next ad.
2. Write down the facts
List what's true about the product: its name, size, colour, what it does and the exact wording on the label. Add any offer you're actually running. Anything you're unsure of stays marked as unknown rather than guessed. This list is your guardrail: every word spoken in the ad and every caption has to agree with it.
3. Pick a format
Three formats cover most product ads:
- Creator-style (UGC): a person holds the product, uses it and says one line to camera. See AI UGC ads for how this format works.
- Demonstration: hands and product, close up, often without a face: pouring, opening, applying, folding.
- Scene-led: the product where it belongs, with a person passing through: a mug on a desk at sunrise, a bag on a train seat.
Choose the format that shows your one idea most directly. If the point is how something looks, feels or sounds, a close demonstration usually beats a talking head.
Two of these in finished, five-second ads: a creator-style clip at a makeup table, and a scene-led clip where the place does the selling.
4. Prepare your references
A video model holds things steady when it's shown what they look like, and drifts when it's only told. You need up to three references:
- The product: a clean, evenly lit photo of the whole product and a close-up of the label. This one matters most; turning a product photo into a video ad covers what makes a good one.
- The person, if there is one: a front portrait and the same person from several angles, in one outfit.
- The place: one image of the setting, plus a short note on the look: the light, the lens, the colour and how polished or raw it should feel.
5. Write the words yourself
If someone speaks, write the exact line. At a relaxed pace people say two to three words a second, and the product needs time on screen too, so a 10-second ad holds one sentence of 8 to 15 words comfortably. Read it aloud with a timer before you use it.
Keep the line to what your facts support. A generated person can say what the product is and what it does. They can't truthfully say they've used it for a month, so don't write that for them.
No dialogue is a fine choice too. Plenty of demonstration ads work best with room sound and a caption.
6. Plan the seconds
Split the ad into stages, usually three: a hook (0 to 3 seconds), the demonstration (3 to 7) and a close with the product clearly in view (7 to 10). For each stage write where everything starts, the one action, where the camera is, where everything ends and what's said. Make each stage's end the next one's start, and the motion stays continuous instead of jumping.
The AI video ad prompt template lays this out in full, with a worked example you can copy.
7. Generate a preview and review it like a client
Watch the first result twice: once at full speed with the sound on, then paused around the moments the product is closest to the camera. Check:
- the product's label, shape, size and colour,
- that the person is the same person from the first frame to the last, with believable hands,
- that the speech is exactly your line, in sync, with nothing added,
- that nothing on screen claims more than the product can back up.
When something is wrong, fix its cause rather than trying again with the same inputs. A wrong label usually means a blurry or missing label close-up. A product that changes size usually means the plan never said how big it is. A blurred, rushed moment usually means a stage asked for two actions at once.
8. Finish it: resolution, captions and a label
Once the preview is right, render the final version at the resolution you'll upload. Add captions timed to the speech, kept clear of the buttons along the bottom and right edge of the app, for everyone watching with the sound off. Label the ad as AI-generated wherever the platform gives you the option.
9. Make three versions, not one
Your first ad is a guess about what will stop the scroll. Make two more that each change one thing, usually the first two seconds: a different opening action, a different line or a different setting. Run them side by side with the same budget and keep what wins. Because AI ads are quick to remake, the advantage isn't any single video; it's how many ideas you can afford to test.
How long it takes
The first ad takes the longest, because that's when you prepare the references. After that, a new ad with the same product and person is mostly writing: a new idea, a new line, a new plan. The generation itself takes minutes.
Making an AI ad in fairytale video
fairytale video follows these steps in order. Upload your product photo and the studio prepares the product sheet for free. Choose a creator from the library, upload a person or generate one, or leave the person out and make the ad with just your product. Then choose a voice and a scene.
Describe the ad in a sentence or two and write the exact spoken lines, or press Write it for me for a first draft of both, which you edit; it never invents prices, offers or claims. Pick 9:16 or 16:9 and a length from 3 to 10 seconds. For each video the studio writes a concept and a timed plan, makes a free preview, and keeps the exact prompt and reference images next to it so you can see what it was given. When a preview is right, render it again with Seedance 2.5 at 720p or 1080p and add captions in the studio.