AI video bytes · pay by the credit
Turn long-form video into bytes that perform.
bytes finds the strongest moments in your podcasts, interviews and streams. You get a frame-accurate editor to finish them: reframe, caption, put titles behind the speaker, export vertical. You approve every frame.
300 free credits, about 100 minutes. Every feature included. Your footage stays private.
“The part nobody tells you about pricing”
“How we got our first 1,000 users”
Built for the people with the most footage to cut
How it works
Full episode to finished bytes. Four steps.
Upload once. bytes takes the first pass.
- SOURCE IN01
Drop a link or a file
Podcast, interview, stream, lecture. Hours long is fine. From a link we fetch only the parts that become bytes.
- ANALYZE02
AI finds the moments
Five passes over the transcript. Every survivor carries a score and the reason for it.
- CUT03
Refine in the editor
Reframe to vertical. Caption to the word. Strike retakes and filler. Put titles behind the speaker.
- RENDER04
Export everywhere
Four aspect ratios. Seven layouts. Most exports render in your own browser, with no queue.
One app, every step
Everything, in one app.
Why we almost shut the company down
Moment detection · ranking · evidence
Every byte comes with its reasoning.
Digest the whole video. Detect in chunks. Snap to sentence boundaries. Re-read every candidate. Re-rank and drop near-duplicates. What survives keeps one score and one sentence saying why.
Hooks aren’t guessed. Each is matched to an archetype measured on real short-form performance. You see the sample size and the median views behind it.
Dozens of candidates, ranked · unsafe ones screened out before they reach you.
Why: A number lands in the first four seconds. The answer resolves inside the clip.
Text behind subject · alpha matting
Put the words behind the speaker.
A real per-frame alpha matte. Not a rectangle, not a blur. Titles and motion graphics sit behind the subject, with the person composited back over the top. One switch.
Per-frame alpha · plate → text → cutout, in that order · every frame of the byte.
Frame-accurate editor · captions · motion graphics
A precise editor, open to everyone.
Keyframe the crop so it follows the action. Split, trim, stitch. Eight caption looks, word-level karaoke, your own emphasis colour. Import and export SRT. Filler, silence and retakes arrive pre-struck in the transcript. You choose what lands.
Ten kinds of motion graphic, written straight into the timeline: quote cards, stat reveals, kinetic lines, callouts, lower thirds, chapter cards, list reveals, big numbers, progress rings, hook titles. Each lands as a normal layer. Move it, retime it, delete it.
Keyframed reframe · SRT in and out · RTL captions · what-you-see-is-what-you-render.
Emphasis words tagged for animation. Filler, silence and retakes offered as micro-cuts.
Export · aspects · layouts · audio
Renders in your browser, not in a queue.
If your machine can handle it, the export runs in your browser. No queue. Anything heavier — a subject matte, a browser missing the codecs — falls back to our renderers on its own.
Music ducks under speech. Gain automation rides the rest. Server renders master to −14 LUFS with a true-peak limiter.
Four aspect ratios · seven layout presets · MP4, H.264 + AAC.
Speaks your language
Three speech engines, routed per language. Nine Indic languages. The language is read from the audio. Garbled passages re-run through another engine.
Windowed ingest
Audio first. Video only around the moments that become bytes. On a six-hour VOD, 95% of the file was never downloaded.
Brand templates
Save your aspect, captions, fonts, logo, CTA and music once. Every new byte opens wearing it. Editing the template never touches bytes you’ve already cut.
One upload, a week of content
One transcript, six formats: blog post, thread, newsletter, LinkedIn post, show notes, YouTube description. In your brand voice. Chapters are one more click.
A/B hook variants
Three alternate openings per byte. Each carries its archetype and the measured lift behind it. Test the hook instead of arguing about it.
Captions in another language
Translate the captions in place and ship the same byte again. Right-to-left scripts lay out correctly in every preset.
AI does the heavy lifting. You make the calls.
One-click tools hand you cuts you have to fix. bytes does the tedious 90%. You do the 10% that decides whether it lands.
Pricing
Credits. Two meters. No subscription.
100 credits = $1. One credit, one cent. Nothing is behind an upgrade — every feature on this page is on for every account. Credits go two places.
Start with 300 credits, free.
That’s about 100 minutes of source, or 30 exports, or a mix: 60 minutes plus 12 bytes. No card. No monthly reset.
Card payments aren’t live yet. Email us and we add credits by hand. That is how everyone gets them today.
Top-ups, once checkout opens
Same rate at every size.
Your footage is yours. We process it to make your bytes. We never train on it and never share it.
FAQ
Questions, answered.
Not yet — publishing and scheduling are on the roadmap. Today you export per-platform 9:16 files and post them yourself.
No. Your footage is yours; we process it to produce your bytes and don't train on it or share it.
Always. The AI ranks candidates; you add, remove, retime, and re-cut. You have final say on every frame — that's the point.
Up to 10 hours and 30 GB per file — one ceiling, the same on every account. Chapter markers for a long episode are a button you press when you want them; nothing is generated behind your back.
MP4, MOV, MKV, and WebM via direct upload — or paste a link from Twitch and other hosts, for content you own. YouTube currently blocks automated downloads from cloud servers, so for a YouTube video download it from YouTube Studio and upload the file.
Three speech engines — Amazon Transcribe, Deepgram nova-3 and Groq Whisper — routed per language instead of one model for everything, with nine Indic languages (Telugu, Tamil, Hindi, Kannada, Malayalam, Bengali, Gujarati, Punjabi, Marathi) explicitly routed. The language is identified from the audio itself rather than guessed, and a garbled pass is automatically re-run on a stronger model. Captions can then be translated in place.
No. If you can trim a video, you can use the editor. The precise controls are there when you want them.
Credits, and nothing else — no subscription, and no feature behind an upgrade. A new account starts with 300 credits and no card. Ingest is 3 credits per minute of source, charged once when the transcript lands; each export is 10. At 100 credits to the dollar, a 40-minute episode costs 120 credits ($1.20) to ingest however many bytes you cut from it, plus 10 credits ($0.10) for each one you export. Paid top-ups aren't switched on yet.
Make your first byte today.
300 credits are already on the account. Drop in a video and see what bytes finds.