BusinessRnD · draft v2 · built 2026-10-21 06:15 UTC
I10 · Product media for merchants, better variations — deep dive
Side bet
Shopify and Google give AI media away. ProofCut avoids that but climbs slowly.Realistic: $0.5–4k MRR in year one · Chance of $10k/mo: low
Cards have Go deeper, Maybe and Kill buttons. Press Send feedback to save. Your marks from all tabs collect at the end of each page.
Prepared 2026-10-20 from Shopify App Store category, app and review pages fetched that day. Review counts are only a rough stand-in for demand; Shopify doesn't publish installs. Notes: research_notes/Owned ideas deep dives/i10-productmedia_notes.md.
Bottom line
Recommended: I10.1, a "proof-first" ad studio on Shopify (working name ProofCut). It uses your ad intelligence to show which ad formats are scaling in a merchant's category, then builds ready-to-run vertical ads only from that merchant's real photos and reviews. No AI model runs when merchants use it, so no billing approval is needed and each ad costs almost nothing.
Generic AI product photos and videos are a dead end. I sampled 47 AI photo, video and ad generator apps: 41 launched in 2025–26, and the median has 0 reviews. Only 4 have 20 or more. The best pure-AI image app, SellerPic, has 90 reviews after 22 months (SellerPic).
Shopify and Google now give generation away free. Shopify's Tinker app bundles 100+ AI tools, including Veo, Kling and Sora video generators and a "Social Media Ad Generator" (Shopify). Google Ads' Asset Studio "turns your website URLs and photos into high-quality videos" (Google). That rules out the Google-video variant and most of the AI-UGC one.
Merchants' top complaints can be designed away: distorted products, "around 40 % of credits" lost on errors, early paywalls and samey posts (SellerPic reviews; Outfy reviews). Real photos, deterministic renders, free re-renders and proven formats answer all four.
Money (estimate):
Pricing: a real free plan, then $19, $49 or $99 a month through Shopify billing. You keep 100% of your first $1M, minus 2.9% processing (Shopify).
Target: about 380 paying merchants at a $30 average gives about $10k/month profit.
Realistic first year: $0.5–4k MRR.
Drop the Google Sheets half of the original idea. Workspace has no built-in billing, Paddle's acceptable-use rules have already blocked marketing tools for you, restricted scopes can trigger yearly security reviews, and no evidence shows buyers are there.
Kill test: about 3 weeks of AI build, then 28 days after App Store approval. You need at least 40 installs, 4 paying merchants, and 25% of installers making 3 or more ads. Otherwise, kill it.
The market today
The Shopify App Store's "Image editor" category lists 406 apps, and "Video and livestream" lists 311 (Image editor; Video). Older template and auto-post tools hold real review counts; new AI generators have almost none; shoppable-video widgets only embed existing clips.
The 2026 wave. 21 sampled generators launched in 2026 alone, such as Klaket (15 Oct), OnPromptAds (7 Oct) and ReelProof (reviews into AI video ads). Almost all have 0–2 reviews at $19–249 a month. Ad intelligence inside Shopify is nearly empty: Meta Adspy has 1 review and Upspring has 8.
What merchants complain about (1–3★ reviews of the leaders):
The AI redraws the product. One merchant says "all it ever did was distort all of your product labels" (Photoroom).
Credits and money are lost. Merchants report losing credits on errors, paywalls before any trial, and surprise annual charges (SellerPic; Minta).
The output doesn't sell, or doesn't finish. Posts "don't seem to engage people enough to get them to make purchases" (Outfy). Renders get stuck with "no progress indicator" (Vidify).
The bundling threat:
Shopify's file editor changes product photo scenes from a text prompt, free on all plans "for a limited time" (Shopify).
Tinker is free, with "generous monthly usage limits". It includes Product Lifestyle Shot, Social Media Ad Generator, Selfie Vlog Generator, and Veo, Kling and Sora video tools (Shopify).
Sidekick generates images, and merchants can connect ChatGPT, Claude or Gemini to their store (Shopify).
Google Ads' Asset Studio makes videos from a URL and photos (Google).
Meta and TikTok likely have similar in-house tools (general knowledge, not re-verified).
What this means: anything a model can make from a photo in one click, the platforms will give away. They don't tell a merchant which ad to make this month, and they don't guarantee the product isn't redrawn.
Variants compared
Variant
Demand evidence
Competition
Fit with your assets
Compute
Verdict
I10.1 Proof-first ad studio
Medium: buyers pay for ad intel; template video apps lead their segment
Medium
Very high
No model
Build
I10.2 AI UGC video ads with proven hooks
Weak: best app has 17 reviews
Very high, bundled
High
Paid provider
Fold into I10.1 later
I10.3 Winning-format feed for agents
Strong for raw ad data on Apify
Medium
High
No model or customer-side
Side product of I10.1
I10.4 Creative-refresh watcher
Indirect
Low
Medium
No model plus renders
Phase 2 upsell
I10.5 EU-localized product media
Weak
Medium
Low
Paid provider
Make it a setting
I10.6 Google Shopping and YouTube video
Need is real but Google fills it
Very high, bundled
Medium
Paid provider
Drop
quadrantChart
title I10 variants by competition and fit
x-axis Low competition --> High competition
y-axis Weak fit --> Strong fit
quadrant-1 On brand but crowded
quadrant-2 Best bets
quadrant-3 Side bets
quadrant-4 Avoid
V1 Proof ads: [0.42, 0.88]
V2 AI UGC ads: [0.83, 0.76]
V3 Format feed: [0.6, 0.7]
V4 Refresh watcher: [0.28, 0.58]
V5 EU localize: [0.45, 0.28]
V6 Google video: [0.8, 0.4]
I10.1
Proof-first ad studio (ProofCut)
What it is: A Shopify app built around a monthly board of ad structures scaling in your category (e.g. "problem, proof, offer, 12 seconds, review quote at second 3"). Pick one, and the app fills it with your real photos and best real reviews, then renders four vertical variants in about a minute for Meta, TikTok, Shorts or the product page. No AI redraws your product.
Who pays: Small and mid-size Shopify merchants who run, or want to run, Meta and TikTok ads but have no creative team. Shopify bills them.
Evidence of demand:
Off Shopify, buyers pay $49–459 a month for Foreplay and $129–959 for Atria (Foreplay; Atria).
Apify's Facebook Ads Library scraper has 45,038 users, 8,994 of them in the last 30 days (Apify).
Template video makers lead creation apps on reviews (Outfy 572, Minta 325), and their bad reviews ask for what this adds: variety that sells.
Weak spot: ad-intel apps inside Shopify have barely any reviews, so demand inside the store is unproven.
Why we can win:
You already have an ad-library collector and a beat-synced renderer.
Rivals are AI generators (fidelity and credit complaints) or template tools (sameness complaints); free platform tools never say what to make.
Using real, attributed reviews keeps US merchants clear of the FTC's ban on AI-faked testimonials (fines of about $51,744 per violation, per Sidley).
Build effort (AI): About 3 weeks of Claude Code work (estimate). That covers the app (from Shopify's official template, so Bettter's skeleton and its ownership question aren't needed), 12–20 video templates, the format-library pipeline and Shopify App Pricing. After launch, the AI refreshes formats weekly and drafts support replies.
Compute: No model at runtime. Videos render on your server at about 1 CPU-second per video second (your measured MontageRnD speed). Claude Code curates the format library offline as operator work on your subscription, at the same cost for 10 or 10,000 merchants.
For $10k/month profit: about 380 paying merchants at a $30 average (about $26.6 net each after the 2.9% fee, servers and roughly 5% tax), or about 3,800 installs at an assumed 10% conversion.
Realistic first 12 months: $0.5–4k MRR (estimate).
First step: I build the format board for skincare, pet and home, plus 20 sample ads on a development store with fictional products. Then the app.
I10.2
AI UGC-style video ads with proven hooks
What it is: Candidate 1. A product page goes in; hooks come from your ad intel; AI avatar or motion clips come out as Meta and TikTok ads.
Who pays: The same merchants, plus a usage charge per generated clip.
Evidence of demand: Weak. Of about 15 such apps launched in 2025–26, the best (Lensia) has 17 reviews; most have none.
Why we can win: Your pipeline is good, but Tinker gives away Veo, Kling, Sora and an ad generator. Avatar "customers" posing as real buyers also run into the FTC rule.
Build effort (AI): 3–4 weeks, plus heavy support on bad generations.
Compute: An approved paid provider at $0.25–1.21 per 5-second clip (fal), passed to merchants as Shopify usage charges. This needs your approval.
Pricing & revenue estimate: $29–99 a month plus about $0.60 per clip. Free redos of failed clips thin the margin; it earns less than I10.1 at any install count (estimate).
First step: Don't build it as a standalone app. Add "AI motion shot" as an optional upgrade inside I10.1 once that passes its kill test.
I10.3
Winning-format feed for agents (Apify actor and MCP)
What it is: Candidates 4 and 5 combined. You sell I10.1's format library and longest-running-ads data to AI agents, agencies and developers, as an Apify actor and an MCP endpoint. Any generation uses the caller's own model or API key.
Who pays: Agencies, media buyers and developers' agents. Apify bills per result and pays you.
Evidence of demand: Strong for raw ad data (Apify ad-library scrapers have 45k, 4k and 3.4k users); weak for generation (the top AI video-ad actor has 148) (Apify store).
Why we can win: Pre-tagged formats beat raw scrapes and reuse I10.1's pipeline for free. But at least seven scrapers compete, and this overlaps I1 and I5.
Build effort (AI): About 1 week on top of I10.1.
Compute: No model at runtime (tags are computed in advance). Any generation runs customer-side.
Pricing & revenue estimate: Pay per result. Expect hundreds of dollars a month, not thousands: the average Apify developer earns about $300–470/month (AgentByline).
First step: Publish it as I10.1's data twin once the library exists, and count it under I1.
I10.4
Creative-refresh watcher
What it is: Candidate 6. The merchant connects their Meta ad account. When an ad tires (frequency up, click-through down), the app renders replacements in the same winning format. Detection needs no model.
Who pays: Merchants spending $3k or more a month on Meta, at $99–299 a month.
Evidence of demand: Indirect. Creative fatigue is a known pain, and I found no Shopify app that handles it.
Why we can win: It keeps customers subscribed. But reading other businesses' ad accounts likely needs Meta App Review and Business Verification (general knowledge, not re-verified): one more platform that can say no.
Build effort (AI): 4–6 weeks, plus Meta's review time. You handle verification.
Compute: No model for detection; replacement ads reuse I10.1's renders.
Pricing & revenue estimate: $99–299 a month from fewer, bigger customers. About 60–100 of them would make $10k (estimate).
First step: Offer it later as an upsell to I10.1 Growth and Pro users. Start Meta verification only if I10.1 passes.
I10.5
EU-localized product media
What it is: Candidate 3. Translates text in images, captions and video overlays into about ten EU languages.
Who pays: Merchants selling into several EU markets.
Why we can win: Hardly; machine translation is a commodity.
Build effort (AI): 2–3 weeks.
Compute: An approved paid provider for translation and image-text editing.
Pricing & revenue estimate: $9–59 a month. Small.
First step: Make language a setting inside I10.1 instead. Template text is easy to localize, and Minta users complained about English-only videos.
I10.6
Video for Google Shopping, Performance Max and YouTube
What it is: Candidate 2. Auto-made product videos for Google campaigns that have none.
Who pays: Merchants running Google Ads.
Evidence of demand: The need is real, but Google now fills it itself: Asset Studio turns URLs and photos into videos with Gemini Omni (Google).
Why we can win: We can't beat free, built into the merchant's own ad account.
Build effort (AI): 2–3 weeks.
Compute: A paid provider or templates.
Pricing & revenue estimate: Low, because Google bundles it.
First step: None. I10.1 exports a 16:9 version for YouTube anyway.
The recommended variant
ProofCut (working name): ads based on what's working now, made only from your real product.
flowchart TD
A["Merchant finds app in App Store search"] --> B["Installs free plan, no card"]
B --> C["Category detected from product types"]
C --> D["Winning formats board for this month"]
D --> E["Picks a format"]
E --> F["Real photos and top reviews filled in"]
F --> G["Edits hook line, previews 4 variants"]
G --> H["Server renders in about a minute"]
H --> I["Downloads for Meta, TikTok, Shorts"]
H --> J["Adds video to product page"]
I --> K["Hits free limit, upgrades via Shopify billing"]
K --> L["Monthly format drop brings them back"]
L --> D
The first three screens (all three are in the mockup):
Winning formats. Your category is detected from your product types. Each card shows a format's structure (hook, proof, offer), length, how many ads use it and how long they have run, refreshed monthly. Abstract diagrams and statistics only; competitors' videos are never re-hosted.
Build. Pick a product. The app pulls its photos and best 4–5★ reviews from Shopify or a review app such as Judge.me (48,770 reviews on its own listing — Judge.me), suggests hook lines from a curated library and previews four variants in a phone frame.
Exports. A progress bar, downloads in 9:16, 4:5, 1:1 or 16:9, "add to product page", and a usage meter with a plain promise: failed renders never count.
Pricing:
Plan
Price
What you get
Free
$0
3 ads a month, small watermark, 1 category
Starter
$19/mo
30 ads, no watermark, 1080p, 1 category
Growth
$49/mo
150 ads, bulk by collection, product-page upload, 3 categories
Pro
$99/mo
500 ads, 10 categories, monthly format report
Add-ons, later
Usage
AI motion shot about $0.60 per 5 s; AI voiceover about $0.20 per ad. Billed by Shopify under a cap the merchant sets (Shopify). Needs your approval
Monthly billing is the default; annual only as an explicit choice (SellerPic's surprise annual charge draws 1★ reviews).
Launch path:
Store: Shopify App Store only, with the required Shopify billing (Shopify).
Listing keywords: "video ads", "UGC ads", "ad creatives" and "product video", plus the nearly empty "winning ads" and "ad library".
Demo: 20 sample ads from a development store with fictional products, posted by the AI to your own Shorts and Instagram through official APIs. No paid traffic.
Reviews: ask in neutral wording only; Shopify bans review incentives.
Kill test: see the go and kill thresholds under Next steps.
Prefer Meta's official Ad Library API; scraping has terms risk
Tag formats; write the hook-line library
Claude Code subscription — operator
You (subscription)
Batch job on your schedule, never triggered by a merchant
Detect category; pick the best reviews
No model
Nobody
Rules on product type, star rating, length and keywords
Fill template and render video
No model
You: server CPU, well under $0.01 per ad (estimate)
Deterministic; product pixels are never redrawn
Rewrite hooks in the merchant's voice (optional)
Customer-side
Merchant
Their own Sidekick or connected AI; copy and paste in version 1
AI motion shot or voiceover (optional, later)
Approved paid provider
Merchant, via Shopify usage charges
$0.25–1.21 per 5 s clip (fal; Kie); needs your approval; merchant's own API key is the alternative
Support replies
Claude Code subscription — operator
You (subscription)
Drafted from a knowledge base; refunds escalate to you
Next steps
gantt
title I10.1 ProofCut launch plan
dateFormat YYYY-MM-DD
axisFormat %d %b
section You once
Partner account and payout setup :a1, 2026-10-21, 3d
section AI build
Seed format library for 3 categories :b1, 2026-10-21, 10d
Video templates and renderer :b2, 2026-10-23, 10d
Shopify app with billing and listing :b3, 2026-10-28, 11d
section Review
App Store review of unknown length :c1, 2026-11-09, 14d
section Test
Demo ads on your own channels :d1, 2026-11-02, 21d
Kill test after approval :d2, 2026-11-23, 28d
Go or kill decision :milestone, m1, 2026-12-21, 0d
I do:
Build the format library (30 formats across skincare, pet and home), the templates, the app, its pricing plans and the listing.
Make 20 demo ads on a development store with fictional products, and post them through official APIs.
Handle App Review, run the kill-test dashboard, refresh formats weekly and draft support replies.
You do once:
Shopify Partner account, $19 App Store registration, payout and tax details (Hyperwallet, USD).
A Meta developer account with identity confirmation, for the official Ad Library API.
Approve the app name and listing text; later, only if the test passes, decide on a paid provider for add-ons.
You do recurring:
Close to zero: refund or legal-sounding escalations, moving payouts into your FOP account each quarter, and a 5-minute look at the weekly go or kill note.
Go and kill thresholds (28 days after approval):
Go: at least 40 installs, at least 4 paying merchants, and at least 25% of installers making 3 or more ads.
Iterate once: 20–39 installs, or payers but weak activation: fix listing and onboarding, rerun 21 days.
Kill: fewer than 20 installs, or no paying merchant by day 28.
Risks and open questions
Demand inside Shopify for "proven formats" is unproven (the ad-intel apps there have 1 and 8 reviews). The kill test is the evidence.
Timing: review time isn't published (Shopify); approval may land near Black Friday (27 Nov 2026).
Data and copyright: scraping Meta's web Ad Library risks breaking Meta's terms, and the official API's coverage of non-political ads outside the EU is unconfirmed. Show abstract formats and statistics only; never re-host other brands' ads.
Music: tracks licensed for apps and ads are needed. Suno's commercial terms are unverified, and the cost wasn't researched.
Platform absorption: Shopify could add "templates from winning ads" to Tinker. The moat is a fresh, category-specific library, which platforms rarely curate.
Where operator work ends: format tagging is a batch job on your subscription, never triggered by a merchant. Confirm you count it as operating.
Unverified: Hyperwallet payouts to a Ukrainian FOP (likely), and whether App Review accepts merchants paying an AI provider directly with their own key.
Reuse: MontageRnD's demos and the VortexMediaBuyingRD radar were built for Vortex, so rebuild them as generic tools. GetHookd data can't be resold.