The Wrapper Test: Why $40 AI Tools Belong Backstage, Never in Your Proposal
Four questions I run before adopting any AI tool: does it have an API, what does a usable asset really cost, would I put a client face in it, and would I write its name in a proposal? DaVinci AI as the case study.
An AI wrapper — a service that resells someone else's models behind its own interface and credit system — is worth paying for when it saves your team time on work nobody sees, and dangerous the moment it becomes something you sell. The test is not "is it good?" It is "does it have an API, and would I write its name in a proposal?" If the answer to either is no, the tool belongs backstage.
I ran that test on DaVinci AI this week. It is a clean example of the category, and the verdict was split in a way that is worth writing down.
What is an AI wrapper?
A wrapper does not train models. It buys API access to models built by other companies — OpenAI, Google, Kuaishou, ByteDance, Black Forest Labs — puts a friendly interface on top, and sells you credits.
That is not an insult. Wrappers do real work: one login instead of five, one bill instead of five, a consistent editor, mobile apps, sensible defaults. For a small team, that convenience is worth money.
But the business model has a shape, and the shape produces the same strengths and the same weaknesses every time. Once you can read the shape, you can evaluate any wrapper in about twenty minutes.
The case study: DaVinci AI
DaVinci AI (not to be confused with Blackmagic's DaVinci Resolve — the name collision is real and your clients will hit it) sells access to a large catalogue of current image and video models through one subscription: Sora 2, Veo 3.1, Kling 3.0, Nano Banana Pro, Seedream, Flux.2, Grok Imagine. It adds an editor — object removal, background swap, upscaling, auto-resize for platform aspect ratios — and iOS and Android apps.
Published pricing on the site: $39.99/month for 5,000 credits (roughly 2,500 images or 125 videos), $99.99 for 15,000, $199.99 for 40,000.
Third-party reviews from the same period quote a completely different ladder — $9, $17.40, $49.80. Both cannot be right. That gap is the first useful data point: the price of a wrapper is a promotion, not a number. Never quote it in a document that outlives the week.
Why the sticker price is not the real price
Credit systems hide their true cost in a place the pricing page never mentions: failed generations still burn credits.
With current image models, a realistic hit rate for a specific brief — right composition, right lighting, no mangled hands, no garbled text — is one usable asset in four or five attempts. So a plan advertised at 2,500 images is closer to 500 assets you would actually put in front of a client. Video is worse, because each attempt costs more.
This is not a scandal. It is arithmetic, and every wrapper has it. But it means you should evaluate on cost per usable asset, not cost per generation, and that number is usually 4–5× the advertised one.
The Wrapper Test: four questions
These are the four questions I now run before adopting any AI tool. DaVinci's answers are in brackets.
1. Does it have an API?
This is the question that decides everything else. A tool you can call from code enters your automation pipeline and scales. A tool that requires a human in a browser is a manual step forever — it caps at how many hours your team has, and it will never appear in your architecture diagram.
(DaVinci: no public API found. This alone moves it from "platform" to "utility".)
2. What does a usable asset actually cost?
See above. Multiply by your real hit rate before you compare to the alternative.
(DaVinci: cheap enough that even at 5× it beats stock licensing for exploratory work.)
3. Would I put a real customer's face in it?
Check the app store data-safety labels, not the marketing site. DaVinci's Google Play listing declares that data is not encrypted and cannot be deleted, and that photos and videos may be shared with third parties.
That single line settles the matter for anything under an NDA, anything with a client's staff or customers in frame, and anything you would be embarrassed to see leak.
(DaVinci: no. Concepts and b-roll only.)
4. Would I write its name in a proposal?
Here is where reputation matters more than features. DaVinci's Android app carries a 2.9-star rating across roughly 185,000 reviews, and the recurring complaint themes are trial-to-charge surprises, paid subscriptions that still demand extra coins, and refund friction. The iOS listing sits near 4.5. That split is itself a signal.
(DaVinci: absolutely not. See the next section for why this is a business question, not a quality one.)
Backstage and front of house
Every tool a service business touches sits on one side of a line.
Backstage is everything that lowers your cost of delivery. Nobody outside the company needs to know it exists. Your only obligations are that it works, that it does not leak anything, and that you can replace it in a week if the vendor changes the deal.
Front of house is anything the client sees, hears about, or is asked to operate. The bar here is completely different: it has to survive being named, googled, and compared.
A wrapper with a 2.9-star billing reputation passes the backstage bar easily and fails the front-of-house bar completely. Same tool, same day, two opposite verdicts — because the two roles are judged on different things.
Most of the mistakes I see are one line crossed in one direction: a team finds something genuinely useful internally and, in their enthusiasm, puts it in a deck.
When a wrapper is exactly the right call
- Pre-sale artefacts. Mockups of what a client's feed could look like, produced before anyone has signed anything. High persuasion, near-zero cost, and nothing sensitive is involved.
- Concept boards for a strategy meeting. Showing beats telling, and the images are disposable by design.
- Creative variant testing. Running one brief through three different models is dramatically cheaper than testing variants with live ad spend.
- Filler and repair. B-roll, background plates, upscaling a client's old low-resolution photos, removing a stray object from an on-site shot.
- Access to a model you don't have yet. When something ships and you need it today, a wrapper beats waiting on an integration.
The rule underneath all five: low volume, high variety, a human reviewing every output, and no sensitive material.
When it is the wrong call
- Recurring production at volume. The moment a task repeats weekly, it belongs on a direct API where it can be automated. Wrappers are a laboratory, not an assembly line.
- Anything with real faces or NDA material. See question 3.
- Anything requiring an exact logo or exact text in the image. Current models still mangle typography, logos, hands and human proportions. Vietnamese diacritics are especially unreliable.
- Anything that becomes a line item. Which brings us to the last question.
What to tell a client who found it themselves
Sooner or later a client reads about a $40 tool that makes videos and asks the obvious question: why am I paying you?
Do not get defensive, and do not rubbish the tool — they can see it works. Answer honestly:
Yes, that tool is real and it is good. It gives you a model. It does not give you a strategy. It does not come to your location and shoot. It does not plan and optimise your ad budget. It does not post every day and answer your inbox. It does not read the numbers for you. All of that work is still sitting on your desk. That is the part you are paying for.
The distinction is not software versus humans. It is a capability versus a completed outcome. A wrapper sells you a capability and hands you the labour. A service takes the labour away.
If the client genuinely wants to make some of their own content, let them have the tool for internal and quick material while you keep the brand-critical assets. That protects the relationship without conceding the account.
One hard rule: never buy it on their behalf or bill it through your invoice. With a refund reputation like that, every billing dispute becomes yours.
The twenty-minute version
Open the pricing page. Open the developer docs. Open the app store data-safety labels. Open the one-star reviews. In that order — because the vendor's own site will never tell you about the last two, and the last two are where the decision actually gets made.
FAQ
Is an AI wrapper a scam? No. Reselling model access with a better interface is a legitimate business, and the convenience is real. The risk is dependency, not fraud: the vendor controls which models exist, what a credit is worth, and what the price is next month.
Is DaVinci AI worth $39.99/month? For internal, exploratory, non-sensitive creative work, yes — it is cheaper than subscribing to four model providers separately. For client-facing production, brand-critical assets, or anything with real people in it, no.
Why does the API question matter so much? Without an API, a tool cannot be automated, cannot be scheduled, cannot be wired into an approval workflow, and cannot scale beyond the hours your team has. It stays a manual step forever.
How do I calculate the real cost of credits? Take the advertised generations per plan and divide by your actual hit rate — typically 4 or 5 attempts per usable asset for a specific brief. Compare that number, not the marketing number.
Should I recommend AI tools to my clients? Only if you sell software. If you sell a done-for-you service, recommending a tool the client operates themselves hands your competitor their entire argument. Answer honestly when asked; do not volunteer it.
Reference: DaVinci AI · CritiquePlus review · Techloy review
#AI #AITools #TechStrategy #AgencyOps #GenerativeAI #BusinessStrategy #Wrappers
✍️ The Author: Do Ngoc Hoan Founder of CookConnects.ca & Wizy.ca. Bridging the gap between advanced algorithms and business execution. I write for technical founders looking to scale their impact with AI and robust engineering.