Everyone's chasing AI video. The boring $9/mo idea is the one that pays rent.

Seventeen new cards today and 57 signals, and the top of the list is exactly what you'd expect: a 75-point AI video tool. Drive any character's motion from a single reference clip — the pitch is "I want my character to move exactly like my reference clip, but every AI tool mangles the face and I keep burning credits."

I get it. That complaint is real, and it's specific, which is rare. Face identity preservation across motion transfer is genuinely the thing that breaks. But here's my problem with it: you're not building a product, you're building a GPU bill with a login page. The people who solve this are running diffusion research teams with eight figures. You'd be reselling someone else's model with a nicer upload flow, and when the next model drops and does it natively, your wrapper evaporates overnight. Credit packs plus subscription sounds great until you do the math on inference cost per generation versus what a hobbyist will pay. Churn on these tools is brutal — people burn $20 of credits in a weekend, get one good clip, and never come back. I'd skip it. If I absolutely had to play here, I'd build the boring part: batch processing and asset management for small studios who generate 200 clips a week. The model is not the moat. The workflow around it is.

The one that actually made me stop scrolling is at 73, and it's deeply unsexy: Create orderable product catalogs for wholesale buyers. The quote is "I send PDF catalogs and buyers manually fill orders—it's slow and error-prone." That is a person describing their Tuesday. That's not a made-up pain point from a brainstorming doc, that's someone with a real spreadsheet and a real margin problem.

This is the one I'd build. Wholesale buyers ordering off a PDF is 1998 behavior and it's still everywhere in furniture, apparel, food service, promotional goods. The buyer doesn't want to log into your SaaS, sure — but the seller will pay $99-299/mo to stop re-keying orders and stop shipping the wrong SKU. Low churn because it becomes their order system, not a toy. The catch: this is a sales-led product. You have to cold call distributors and get on the phone with a 58-year-old who runs a showroom. Most indie hackers won't do that, which is precisely why the opportunity is still sitting there at 73 instead of being a YC company. Stack: Next.js, Postgres, PDF parsing for the catalog import (that part is genuinely annoying and where I'd spend the first week), and a shareable order link with no login required for buyers. Stripe for the seller's subscription only.

Third one, 71: SaaS subscription ledger with renewal and waste alerts. Free up to 5 subscriptions, $7-9/mo personal, $19-29/mo team. I want to like this more than I do. The "we found $340/mo of zombie subscriptions" alert is a genuinely good demo, and the number is the marketing. But this lane is crowded — Rocket Money owns consumers, Cledara and friends own teams, and every bank is bolting on a subscription view. Your wedge has to be the team/shared-card case, where nobody actually knows who owns which seat. That's a real gap, but it's a hard one to sell from a cold start because it needs company card data. I'd pass unless I already had an audience of ops people.

Worth noting what didn't make my shortlist: speech-to-text API is expensive and hard to integrate landed at 72, and that's a price war I have no interest in entering. Whisper is free. You will not out-cheap free.

The pattern I keep seeing: the highest score goes to whatever sounds most futuristic, and the money goes to whatever sounds most boring. Seventeen cards, and the only one I'd actually start tonight involves PDFs and phone calls.