Adverts: what the competition runs, and how we make ours
The short answer
You are right that our reels are not good enough for the test. But the fix is not a "professional" advert. The adverts that win in our market look like a parent's phone video: real hands, a real book, a voice, big captions, a child's face. We can make that with what we own, mostly for free, and automate it per book.
- What Wonderbly runs (51 adverts live in the UK today): 5 to 12 second vertical videos filmed on a phone. Hands open the box, show the cover with the child's name, turn a page. A voice tells a one-line story. Captions are burned in. Some are reposted customer photos of a child hugging the book. Not one looks like a TV advert.
- What our reels are: four still pictures with captions and music. Nothing moves, nobody speaks, there is no real book and no child's reaction. The paintings are good; the format is the problem.
- AI actors: they exist (HeyGen, Arcads, Creatify) and an AI presenter as "our voice" is legal. An AI "mum" saying she bought it is a fake testimonial, which is banned in UK law and against Meta's rules. And on a likeness product, an AI person invites viewers to judge our AI. Not for the first test.
- What I recommend: three adverts, two of them fully automatic from the files we already have, one with real hands on the real book (20 minutes of filming once, or AI-generated hands from our own cover photo at about £3 a clip). Details, prices and the legal lines below.
1. What the competition is actually running
Read today in Meta's Ad Library, United Kingdom, active adverts only.
| Who | Live adverts | What they look like |
|---|---|---|
| Wonderbly | 51 | Phone-filmed unboxings, 5 to 37 s (most 9 to 12 s), voice + captions, one book per advert, Christmas and sibling titles, reposted customer photos, "trusted by over 10 million people" |
| Librio | 23 | 26 s videos, one niche per advert (bilingual families), "order now" |
| Hooray Heroes | new set 9 Oct | Still images, emoji bullet lists: "3 million books sold", "illustrated by real artists", "made and shipped from the UK" (aimed against AI books) |
| Magical Children's Book (AI, from a photo) | 8 | Still images, "illustrations made from their photo", child and dog in one book. Our closest advertising rival, 1 review on Trustpilot |
| My Story Tale Books | 3 | Still image, hook: "THEIR FACE WHEN THEY REALISE THE STORY IS ABOUT THEM" |
| Story Bug | 2 | Still images, five stars, "#1 rated", "gift for your child or grandchild" |
| Booklydoo (name + photo) | ~25 | Still images, tick lists, "4.1/5 on Trustpilot", "all it takes is one photo" |
| Tickled Moon | 1 video | 12 s, the hook is a review: "She saw her name on the first page and her eyes went wide" |
| I See Me! (US) | 3 | 22 s video, Christmas, "save 35%" |
| MyLittleStories | 0 | No adverts. They grow on Trustpilot bursts and ChatGPT |
Wonderbly's own frames, as they run today
The pattern, in one line: real hands + real book + a voice telling a one-line story + big captions + the child's reaction + stars or "millions sold" + a Christmas reason to buy now. Nothing is polished. The amateur look is the format, because on Reels a glossy advert reads as an advert and gets scrolled past.
2. Our reels next to theirs, honestly
- What is good: the paintings. The "same photo, now painted" idea is a stronger hook than anything Wonderbly has, because they cannot show it.
- What is missing: motion (nothing moves), a voice, a real printed book in real hands, a child's reaction, and any proof (stars, reviews, "X books sold"). Also the caption says "painted by hand", which the Trustpilot rule of 5 September already banned: it must be "painted in our studio".
- What this means for the test: if we ran these as they are and got nothing, we would blame the channel for a format problem. You were right to stop.
3. The tools, what they do, what they cost
£ at 0.79 to the dollar. An "8 s clip" is one shot of the kind Wonderbly cuts together three of.
A. Make video from our own pictures (image to video)
| Tool | Cost | Good for | Catch |
|---|---|---|---|
| Google Veo 3.1 Fast we have the key | £0.76 per 8 s clip, 1080p, with sound | Start from our real cover photo, "hands open the book on a kitchen table". Takes up to 3 reference pictures, so the cover stays ours. | Fingers and the printed text can go wrong; expect 2 to 4 tries per usable clip, about £3 each. Veo 3 (older) is deprecated; 3.1 is current. |
| Gemini Omni Flash same key | about £0.63 per 8 s | Google's new default video model, cheaper | Newer, less proven for products |
| Kling 2.5 Turbo | about £0.44 per 8 s, no sound | Cheapest motion | Needs a reseller account (fal is retired for us) |
| Runway Gen-4.5 | £0.76 per 8 s, no sound | "Object locking" keeps the product shape | Picture only, 720p |
| OpenAI Sora 2 | - | Shut down 24 Sept 2026 | Not an option |
| Meta's own tools in Ads Manager | £0 beyond the ad spend | Turns up to 20 product photos into a multi-scene video, animates stills, adds music and text | Generic look, Meta adds an "AI info" label itself. Good for extra variants, not the hero |
B. AI presenters and "AI actors"
| Tool | Cost | Good for | Catch |
|---|---|---|---|
| HeyGen | free (3 videos a month, watermark), then $29 a month | A presenter who talks; "product placement" puts a product in their hand | Their own notes say it works best with small items (cosmetics, gadgets). A hardcover may come out scaled down |
| Arcads | about $110 a month for about 10 videos | The most realistic "person talking to camera with a product" adverts | No free trial. Reviews are mostly from sellers |
| Creatify | $33 to $49 a month | Paste the product page, get an advert with a presenter | Lower realism; the book is "around" the presenter, not the star |
| Captions | $24.99 a month | Editing, captions, some AI presenters | We already do captions free with ffmpeg |
C. Voice
Kokoro (free, already in your TikTok pipeline) or ElevenLabs (about $5 a month) read a three-line script in a warm British voice. This is the single cheapest upgrade: Wonderbly's adverts all have a voice.
4. The legal lines for AI in adverts (UK, checked today)
- An AI presenter speaking as the shop ("we paint your child from one photo") is allowed. The ASA's June 2026 guidance: the same rules apply however the advert was made; it must not mislead and must not fake an endorsement.
- An AI person pretending to be a customer ("I bought this for my daughter and she cried") is a fake testimonial. Banned under the Digital Markets, Competition and Consumers Act since April 2025, and the ASA calls it a misleading endorsement. Do not do it.
- Labels: Meta labels adverts it detects as AI-made with an "AI info" note in the menu; a photorealistic AI human gets a label next to "Sponsored". There is no UK law that says we must write "made with AI" on a product advert, but a label cannot rescue a misleading one.
- Our own paintings in the advert are the product, shown truthfully. Fine. The caption must say "painted in our studio", never "painted by hand" (your 5 September rule).
- Children's faces: only the demo children we own, or a real customer's child with written consent, first name only. Already the house rule.
5. What we make: three adverts, mostly automatic
Advert 1. "The reveal" £0, fully automatic
Our existing four beats, rebuilt to the winning format: the real photo, then the same picture painted, then the close-up page, then the cover and price. New: every still moves (a slow push-in), a voice reads three lines, the captions are bigger, and the last frame carries the 14 December date. I change the reel tool once (about 2 hours); after that it is one command per book, and the robot can make one for all 17 books overnight.
Advert 2. "The unboxing" the one that needs a decision
This is Wonderbly's format: a box opens, the cover shows the child's name, pages turn, the camera lands on the child's painted face. Two ways to get it, pick one:
- Real hands, 20 minutes, once: the printed Pocketful book (Priya) plus the Christmas proof copy when it arrives. A phone on the kitchen table, hands only, no face, no talking: open the box, lift the book, turn four pages, hold on the face. I cut it, add the voice and captions. Cost £0. Honest view: this is the best advert we can make, because it is the real thing.
- AI hands, no filming: Veo 3.1 Fast starts from our real cover photo and makes the "hands open the book" clip. About £3 per usable clip after retries, so about £10 for the three shots an advert needs. Fully automatic per book. Risk: the printed text or the painting inside the book changes between frames, and I reject the clip. It will look slightly less real than your phone.
Advert 3. "The still" £0, fully automatic
One picture: the photo and the painting side by side, a one-line hook on top ("One photo. Then they are in every page."), the price and the date. Hooray Heroes, Story Bug and Booklydoo run stills like this; they are the cheapest way to try five different hooks and let Meta tell us which one people click.
How they are tested
- Three adverts, two hooks each (the "same face" hook and the "Christmas present sorted" hook) = six variants in one ad set. Meta shows the winners more by itself.
- The pixel from the growth plan tells us where people stop: advert (no clicks), page (clicks, no sample orders), or samples (orders, no payment). So a bad result is never blamed on the wrong thing again.
- A real customer clip beats everything above. The £10-off-for-a-photo offer in the growth plan becomes "a photo or a 10-second clip of them opening it". One real reaction clip is the Wonderbly advert, and it is legal with the family's written consent.
6. What gets automated, and what stays by hand
| Step | Who | Per book |
|---|---|---|
| Pick the close-up page (the face must be big) | me, by eye, as now | 2 minutes |
| Advert 1 and Advert 3, all hooks, captions, voice, music | robot | one command |
| Advert 2 clips from the cover photo (Veo) | robot generates, I check every clip by eye | about £10, 15 minutes |
| Advert 2 with real hands | you film once, I cut | 20 minutes, once |
| Upload to Ads Manager | me, through the Marketing API once the ad account exists | minutes |
| Reading the results | me, on the dates in the growth plan | - |
The 21 September note in the reel tool said paid Facebook adverts were rejected for this Christmas because a click costs about £1 in the Christmas quarter and Meta needs a bigger budget to learn properly. The second point is still true; that is why the growth plan calls this a £250 test with stop rules, not a campaign. Your "yes" today supersedes the 21 September rejection, and the record will say so.
7. What I need from you
- Advert 2: your hands for 20 minutes with the printed book, or AI hands for about £10 of tries? (You can say both.)
- OK to spend up to £20 on Veo clips while I find the prompt that keeps the cover and the painting straight?
- OK to ask the 13 paying families for a 10-second clip as well as a photo, with the £10 off?
- Nothing else. The ad account step from the growth plan stays as it was.
8. Honest odds
- With Adverts 1 and 3 alone (free, automatic): we would match the smaller rivals' quality (Magical Children's Book, Booklydoo), not Wonderbly's. Fair test, not the best test.
- With Advert 2 on the real book: we match Wonderbly's format, and we have the one thing they cannot show, the photo next to the painting. That is the test worth running.
- With one real customer clip: better than anything we can make ourselves. Worth the £10 offer on its own.
- What a good advert cannot fix: the page after the click. One in four sample families pays. The pixel will show whether the advert or the page loses them.
Sources read today: Meta Ad Library, Wonderbly UK and the other advertisers by name; Google Gemini API pricing (Veo 3.1, Omni Flash); Veo 3.1 docs; OpenAI deprecations (Sora 2); Meta on AI labels in adverts; ASA June 2026 guidance, summarised by Lewis Silkin; CAP on AI in advertising; tool prices from HeyGen, Creatify and an independent round-up (blog figures, check before buying); Creative Review on Wonderbly's Facebook approach (older, still true: one book per advert, product shots are the hero, 60% of sales in the last quarter).