Launched Klyping - AI that watches your short-form video before you post and tells you the fix

3 weeks ago I had a messy Gemini prompt and a Next.js boilerplate.

Today there's a live product, paying users, and a Google Ads campaign

about to go live. Calling this my IH launch.

The problem that kept bugging me

Every analytics tool on the market measures what already happened.

Views, retention, drop-off — all POST-publication. Completely useless

if what you want is to know what's wrong BEFORE you publish.

I was posting TikToks and Reels weekly, some hitting 20k views, most

flopping at 400, with zero visibility into why. "Post and hope" isn't

a strategy.

What Klyping does

Upload any short-form video. Gemini 2.5 Pro (multimodal) watches every

frame and hears every word in a single pass. You get back:

- Viral score 0-100 — calibrated brutally honest (average real video

scores 45-60, not 85 like most tools would tell you)

- Hook analysis with first-3-second breakdown

- Exact timestamp where viewers drop off

- Platform-specific tips (TikTok/Reels/Shorts each have different

algorithmic priorities)

- The single Priority Fix — the one change that moves the needle

- Hashtags, best post time, title variants, thumbnail score

Free tier: 3 credits on signup + 1 daily. Paid packs: €5/€10/€20

one-time, no subscription. Credits never expire.

The stack

- Next.js 16 App Router

- Gemini 2.5 Pro multimodal — the hardest part was prompt-engineering

it to NOT inflate scores. Default Gemini is way too optimistic.

- Supabase (auth + Postgres + full RLS + Storage)

- Paddle (Merchant of Record — they handle EU VAT which was going to

be a nightmare otherwise)

- Vercel

- Full i18n: EN/ES/FR/DE/PT including the PDF report

Decisions I'd love pushback on

1. Credits over subscription. Usage is bursty (creators don't analyze

daily), and people hate recurring charges for tools they don't

touch for 2 weeks. Gamble: lower MRR predictability.

2. Free tier uses Gemini Flash, paid uses 2.5 Pro. Quality gap is

real — Flash is noticeably worse at nuanced platform-specific

advice. Is the free tier good enough to convert?

3. Brutally honest scoring. Everyone inflates. I deliberately don't.

Gamble: users might leave disappointed with a 47/100 and not come

back for the fix.

What's shipping next

- Meta Pixel + retargeting

- SEO landing pages for long-tail keywords ("ai hook analyzer tiktok"

etc.)

- Referral ("give 5 credits, get 5")

- Smart prompt tightening to squeeze more quality per token

What I'd love feedback on

- Landing in <5s test: does it communicate the product?

- Credits vs subscription — what would YOU prefer as a creator?

- First-session friction?

- Any similar product I should benchmark against?

Try: https://klyp.ing (3 free credits, no card needed)

Happy to AMA about Gemini multimodal prompt engineering, Paddle vs

Stripe for EU, or why I think subscriptions are dead for creator

tools.