A trend video studio, not an AI circus

AI video generation arrived before AI video usability. The models got good faster than the interfaces did — prompt boxes, token budgets, aspect-ratio jargon and model name soup stood between a person with two photos and a clip they actually wanted to post. AI Gen Video is our answer: every trend is a template with a fixed scene, a fixed camera language and a fixed mood, so the only creative decisions left are the ones that matter — which photos, which scene, how long.

We are deliberately narrow. Four trends, not four hundred — each one tuned until it produces the clip people actually go viral with, instead of a generic text-to-video endpoint with a gallery. That narrowness is what lets us promise a two-minute turnaround, a live price before you confirm, and credits back automatically when a generation fails.

Your photos pass a free readiness check, then an automated safety review. The chosen model renders the scene around your likeness, native audio included where the model supports it. The finished clip passes its own safety review, lands in your library with a download link, and is kept for 14 days — your uploads expire on the same schedule. Credits are held while the job runs and only captured on delivery; if anything fails, they return in full without a support ticket.

The models behind the trends are MiniMax H3 and Seedance 2.5, reached through a redundant provider setup so a single vendor outage never stalls a job. Everything runs over HTTPS with signed, expiring links — your photos are never public pages.

No watermark added to your clips. No hidden per-export fees. Packs, subscriptions and single videos — all pay-as-you-go friendly. Your photos are used for your generation and then deleted on schedule, never sold, never used to train anything. Content is reviewed automatically on the way in and on the way out, and the content policy is enforced the same way for everyone.

No impersonation tools — the line between "a duet with my friend" and "a fake of someone famous" is the whole business, and we stay firmly on the near side of it. No scraped-photo generation — every input must be yours or authorized. No infinite-scroll generation farm — the three-parallel-job limit is deliberate, because quality and moderation both degrade when volume is the only goal.

No dark patterns on pricing either: one price list, live totals before you confirm, automatic refunds on failure, and a 14-day storage window that is stated before purchase instead of discovered after.

The near-term list is concrete: more scenes for the existing trends before any new trend launches, a wider model catalog as the providers we work with prove their reliability, and self-serve tools for the things that currently need a human — bulk photo checks, batch re-renders, and clip trimming before download.

The long-term bet is that templates beat prompt boxes for most people most of the time. The models will keep improving; the interface should keep getting simpler. If a future feature would make the flow harder to explain in one sentence, it waits.