Launching today
Kineo
Turn any article into a cinematic AI video — BYOK
8 followers
Turn any article into a cinematic AI video — BYOK
8 followers
Turn any article into a cinematic short — script, voiceover, AI visuals, fully automated. Bring your own provider key: pay compute at cost, zero markup.
Hey Product Hunt 👋 I'm the maker. Kineo turns an article, URL, or topic into a short AI video — script, voiceover, AI visuals, word-by-word subtitles, and an optional talking-head avatar of you from one photo + a 30-second voice clone.
What makes it different: it runs on your own provider key (BYOK). Compute is billed to your WaveSpeed account at the provider's price with $0 platform markup, and you see the estimate before every video — no bundled credits, no reseller margin.
See real outputs (no signup): https://kineo.studio/examples
Would love your honest feedback — I'm reading every comment.
how does it handle longer pieces like 4000+ word articles, and are the AI visuals consistent across scenes or does it reuse the same stocky look throughout?
@yusufaktulqthz Both good questions.
Long articles: anything over ~8,500 characters goes through a condensation pass before scripting — the article is chunked into sections, each section's facts are extracted in parallel, then recombined (repeating if it's still too long). So a 4,000-word piece isn't head-truncated; the script is written from the whole thing. The real constraint is video length, not article length: you pick 30s / 1 / 2 / 3 min, and the script is written to that budget, so a longer article just gets condensed harder.
Visuals: you pick one visual style (cinematic, documentary, anime, cartoon, 3D, vintage, cyberpunk, minimal) and that directive is applied to every shot's prompt, so the look and grade hold across scenes. Each shot is generated from its own beat but grounded in the whole article, and shot types rotate — never three identical framings in a row.
Being straight about the limit: there's no cross-shot character lock. You get a consistent style, not the same actor in every scene. The exception is the talking-head presenter, which is your own face and stays consistent throughout. And the "stocky look" you're worried about is a real-footage share you control in the shot mix — set it to 0 and every frame is generated. If you pick a stylized look like anime or cartoon, stock footage gets folded into AI automatically, so you never get a real clip spliced into a cartoon.
Real unedited outputs, no signup: https://kineo.studio/examples
How does it handle long-form articles, like 3000+ word pieces — does it split them into multiple shorts or try to cram everything into one video?
@evcenmakbu56049 Neither, really — it's always one video, and it condenses rather than crams.
Anything over ~8,500 characters gets chunked into sections, each section's key facts extracted in parallel, then recombined (repeating if it's still too long), so the script is written from the whole article instead of just its first few thousand characters. Then you pick the length — 30s, 1, 2 or 3 minutes — and the script is written to that budget. A 3,000-word piece into a 60-second video means aggressive condensation, not cramming; you'd want 2–3 minutes if you care about keeping the detail.
There's no auto-split into a series yet. If you want a multi-part treatment today, you'd run it twice with different framings.
Real unedited outputs, no signup: https://kineo.studio/examples
How does it handle the voiceover quality across different languages, and are there limits on video length when using my own API key?
@taha188100 Hey Taha — both fair questions, answering precisely.
Voiceover across languages. Kineo narrates in three languages today: English, Mandarin, and Cantonese (Mandarin in Traditional or Simplified script). I'd rather name three we've actually tested than claim "50+ languages." TTS is MiniMax speech-02-hd, and every generated clip passes an automatic QA gate before it's used: a crest-factor + duration check that catches the clipped, buzzy renders these models occasionally produce, plus a transcript-match pass that re-transcribes the audio and compares it against the intended line. Fail either and it retries, then falls back. You can also clone your own voice. Honest caveat: a cloned voice carries the accent of its reference audio, so it doesn't switch languages cleanly — for cross-language work the preset voices are more reliable.
Length limits on your own key. The cap is ours, not your key's: max 3 minutes (options 30s / 1 / 2 / 3 min). The free plan is 1 minute and 3 videos per rolling 7 days; Pro removes the video-count limit, and the 3-minute cap still applies. Your API key only decides what the compute costs — Kineo adds $0 on top, you pay the providers their real price.
The genuinely interesting part of "3 minutes": three minutes of Chinese is not three minutes of English. Every language, every voice — and every cloned voice — has a different real speaking rate, so we never hand the script writer a plain word count. We measure each voice's actual pace (characters/sec for Chinese, words/sec for English; a voice you clone gets measured from your own reference audio), multiply it by the duration you chose, and give the writer that budget. Segments are planned in the same unit and land in a 6–8 second window. Then at assembly, if a shot and its narration disagree, we time-fit the video to the audio — never the audio — so nothing gets chipmunked. And talking-head segments are never time-fitted at all, because that would break lip-sync.
So "3 minutes" means the finished video lands at ~3 minutes, in whichever language you picked. That calibration is the system's job, not yours.
Spanish and Japanese are the leading candidates for what we add next.
Pasted a long tech article and got a surprisingly watchable 60 second video in under a minute, the voiceover pacing felt natural and the visuals matched the tone.
@sradfzq Thanks for actually trying it — that's the feedback I care about most.
One clarification for anyone else reading, because I'd rather set expectations than oversell: a 60-second video doesn't render in under a minute. Every segment runs its own TTS, image and image-to-video call, and those are 30 seconds to a few minutes each on the provider side, then there's the ffmpeg assembly. Budget a few minutes for a 60-second clip, more on the higher-quality model tiers. If it appeared instantly for you, you were probably watching one of the pre-rendered examples.
Glad the pacing and tone landed — the voice pacing is measured per-voice rather than assumed, which is the part that took longest to get right.