Klyve - Turn a prompt into a motion graphics ad for your SaaS
by•
Klyve's AI director composes a bespoke, on-brand motion graphics ad for your SaaS, pulling in your real colors, screens, and narration. No code, no editors, no timeline. Just a prompt and a finished ad.
Replies
Best
Maker
📌
Hey Product Hunt 👋
I built Klyve because making a decent product ad always meant one of two bad options: spend hours in After Effects yourself, or pay a freelancer and wait days for a first draft.
This is actually version 2. The first version taught me a lot about what breaks and what people actually want out of a tool like this, and I rebuilt a good chunk of the pipeline based on that.
You give it a prompt, your brand colors, and your product screenshots, and its AI director composes a full motion graphics ad end to end. No editing timeline, no code required.
One thing I cared a lot about under the hood: I didn't want the AI writing freeform animation code, since that tends to break in weird, unpredictable ways. So Klyve generates structured output that gets validated against a schema before Remotion ever renders it. That's what keeps the quality consistent instead of random.
I built this solo. Would genuinely love to hear what you think, and I'll be around all day answering questions about how it works or what's coming next.
Report
Validating structured output against a schema before Remotion renders is the right call, but a schema catches malformed, not wrong. A caption that overflows the safe area, or a logo landing on a same-colour background, is perfectly valid JSON and still a bad ad. Rendering three frames and checking contrast and overflow would catch more than the schema does, and it's cheap. The more interesting detail here is that v2 was a pipeline rebuild rather than a feature list.
Replies
Validating structured output against a schema before Remotion renders is the right call, but a schema catches malformed, not wrong. A caption that overflows the safe area, or a logo landing on a same-colour background, is perfectly valid JSON and still a bad ad. Rendering three frames and checking contrast and overflow would catch more than the schema does, and it's cheap. The more interesting detail here is that v2 was a pipeline rebuild rather than a feature list.