What is LuniSpark? (What)
LuniSpark turns a video or podcast transcript in SRT, WebVTT, or plain-text form into highlight quotes and ready-to-edit draft posts for Facebook, LinkedIn, and X. Quotes are rendered word for word from the cues you select, each with its transcript timecode attached.
When should you use LuniSpark? (When)
Use it after publishing a video, podcast episode, webinar, or lecture, when you need social posts that quote what was actually said instead of paraphrasing it from memory or inventing a plausible line.
Where can you use LuniSpark? (Where)
Open /lunispark/ in a modern desktop or mobile browser. The transcript is processed on your own device; nothing is uploaded, and no account is needed to use the beta. Open the tool or explore other Lunibyte guides.
How do you use LuniSpark? (How)
Import the transcript file or paste the text, confirm the content is yours to use, tick the highlight cues (or start from the labelled rule-based suggestions), review the verbatim quotes and their timecodes, then copy the drafts or download the Markdown export with citations. Evaluate the benefits in your own context; a local-first design does not replace your responsibility to review outputs.
Why use LuniSpark? (Why)
It keeps promotion honest and cheap: quotes can never be fabricated because they only render from your transcript, and the AI budget is zero by design because the planned intelligence, LuniSpirit, runs locally on your device instead of renting a hosted model.
Important limitations
The beta is draft-only and never posts for you. Highlight suggestions are rule-based, not AI, until LuniSpirit is ready; files are capped at 1 MB, a malformed timestamp rejects the whole file, and everything lives in memory only, so reloading the page clears the session.
Why LuniSpark? Your Transcript Should Not Need an Upload or a Subscription
Turning an episode into social posts is a small, well-understood job: read a transcript, find the good lines, shape them into drafts. Yet most tools built for it ask for three things that have nothing to do with the job — your file on their server, an account with your content history, and a monthly subscription that mostly pays for rented AI tokens. LuniSpark was designed by refusing all three, and this page explains what you get in exchange for each refusal, and what it costs you in honesty about the beta's limits.
No upload, because your transcript is not ready to be public
A transcript is often the most sensitive version of your content: the unedited episode, the internal meeting, the interview before cuts. Sending it to a conversion or AI service means a copy now lives somewhere you do not control, under a retention policy you did not read. LuniSpark processes everything in your browser. The file is read on your device, held in memory only, and gone when you reload the page — the page even warns you before that happens. You can verify the claim yourself: open your browser's network tools, import a transcript, and watch nothing leave. A tool whose privacy claim survives the network tab is a different thing from a tool with a reassuring privacy page.
No subscription, because the AI should run on your device
The standard pricing of AI content tools is a toll on someone else's model: every summary rents tokens, so every user rents them forever. We are building the alternative into LuniSpark itself. Our local agent, LuniSpirit, will read transcripts and propose highlights entirely on your device — the computer you already paid for, doing work it is perfectly capable of. The AI budget of this product is zero dollars, permanently, and that is a design decision rather than a promotion: when no per-use cost exists, there is nothing to meter, cap, or upsell.
Until LuniSpirit is ready, the beta is transparent about the gap: highlight suggestions come from visible rules — cue length and repeated key phrases — and are labelled "not AI" in the interface. We would rather ship an honest rule than a borrowed model wearing our name.
Verbatim quotes, because trust is the feature
The deepest reason for LuniSpark's design is its quote rule: a quotation is rendered only from a cue you selected in your own transcript, word for word, with its timecode attached. Summaries and templates may surround the quotes, but the quoted words themselves can never be invented, improved, or quietly rephrased. For creators, that is the difference between promotion and misrepresentation. For teams quoting a guest, a customer, or an executive, it is the difference between a draft you can send for approval and a liability you have to fact-check line by line.
The honest limits
A "why" page that hides the trade-offs is just an ad, so here they are. The beta is draft-only: you copy posts and publish them yourself, because connecting your accounts is exactly the kind of access this tool refuses to ask for. Files are capped at 1 MB, generous for captions but not for a raw book manuscript. Highlight selection is manual or rule-assisted today, not yet the local AI we describe above. And because nothing is stored, closing the page ends the session — download the Markdown export if you want to keep the drafts and their citations.
If those limits fit how you already work, LuniSpark is ready: free, in your browser, and on your side of the upload line.
From the Lunibyte team