Want the output, not the API
Creators who read the launch news
The model made headlines; your deliverable is still a subtitled video. Upload, review the cues, export — the engine layer is already handled.
Gemini 3.5 Transcribe Subtitles — Without the API.
Drop your video or audio hereor
MP4, MOV, MP3, WAV, M4A. Up to 2 GB.EnglishFollows source
Google's new model made speech-to-text headline news — as a developer API. If what you want is a subtitled video, upload it here instead: frontier-class engines benchmarked per language, cues timed and checked, SRT or burned-in captions out in minutes. No API key, no code. Free 15-minute preview per file.
No card needed. First 15 minutes free.
How does it work?
Launch-day review
MP4 · 14:02 · uploaded
Upload a file, or paste a public linkAny of 95+ source languages
Timed subtitle cues
00:04
Let’s talk about the new transcription models.
00:09
Accuracy jumped again this year.
00:13
But most of them are APIs, not apps.
Download
launch-review-en.srtlaunch-review-fr.srtall-formats.zipSubtitle file
SRT · VTT
Bilingual subtitles
source + translation, cue by cue
Burned-in video
captions on the frame · up to 4K
Translation
into 95+ languages
Text files
TXT · DOCX · XLSX · Markdown
Nine reasons to upload here instead of wiring the API.
Every row below is a piece of the pipeline you would otherwise build yourself.
Hallucinated lines get caught, not shipped
- Two detectors, not oneAI hallucination detection and engine-side checks run on every chunk, so invented text is flagged where it happens.
- Automatically re-runA flagged segment goes back through a different engine instead of landing in your file.
- Silence stays silentA pause is a pause, not a sentence the model imagined.
The file the platform asks for
- Subtitle filesSRT and VTT, single-language or bilingual.
- Text and tablesTXT, DOCX, XLSX or Markdown, timestamps included.
- One ZIPEvery format in a single download.
- Share linkSend the result to a reviewer — no account needed on their side.
Start from any video you have
- Upload or paste a linkMP4, MOV, MP3, M4A — or a public YouTube, Instagram or Facebook URL.
- Long files welcomeUp to 8 hours and 30 GB per file, on every plan.
- Record in the browserCapture the screen or your voice with nothing else installed.
- Bring subtitles you haveImport an .srt or .vtt and translate or edit from that.
Pick the subtitle languages once — 95+ to choose from
- Any source, any targetTranscribe the spoken language, then translate into as many targets as you need.
- One track per languageEach language becomes its own subtitle track you can open, edit and export.
- Readable line lengthsCues are segmented for reading speed, and lines too dense to follow are flagged in the editor.
- Bilingual pairsSource and translation can ship together in a single subtitle file, cue by cue.
Proofread before anyone reads
- A score out of 100Suggestions sorted by kind, usually in ten to thirty seconds.
- Approve by groupApply the groups you agree with; the rest is left exactly as it was.
- Timings never moveMisheard words and punctuation get fixed without the cues shifting underneath.
- Entirely optionalNothing is applied until you press the button.
Captions that look like your channel
- Two caption stylesOutline and shadow, or a background box with its own opacity.
- Styled per trackPrimary and secondary styled separately, so bilingual subtitles stay readable.
- Adaptive font sizeLong lines stay inside the frame instead of running off the edge.
- Previewed on the frameFont, size, colour, alignment and margin, shown on the real still as you set them.
Need the finished video? Burn it in, up to 4K
- Up to 4K720P, 1080P, 2K or 4K, with the style exactly as you set it.
- Bilingual burn-inTwo languages stacked on one frame, each styled separately.
- Your watermark, or noneHide the default one entirely, or replace it with your own.
- Renders in the backgroundKeep editing; the job uses a snapshot from when you pressed render.
Your names survive the subtitles
- Mark a term universalPeople, products and jargon come back spelled your way in every track.
- Or pin it per languageWhen a market really does use another name, set that rendering just for it.
- A workspace glossaryApplies automatically, on every project you start.
- Import or extractLoad terms from CSV or XLSX, or let the AI pull candidates out of a PDF.
The same pass writes more than subtitles
- Chapters and quotesReal timestamps and the lines worth pulling, ready to paste.
- Description and tagsThe upload copy comes off the same transcript.
- Summaries, translatedThe second market gets more than translated captions.
- Ask the videoEvery answer points back at the cue it came from.
Who lands on this page — and what they actually need.
If the deliverable is a captioned video, the API was never the requirement.
Prototype before you build
Developers mid-decision
See what production-grade cue segmentation, reading-speed flags and bilingual exports look like before committing to writing your own.
Hours of recordings
Course & training teams
Long files process in one pass — up to 8 hours and 30 GB each — and come back captioned, searchable and ready to publish.
More than one language
Global channels
Transcribe the source, translate into any of 95+ targets, and ship bilingual subtitles or burned-in captions per market.
And the rest of it.
None of these are the reason to switch, but every one of them is there when a delivery spec asks for it.
Upload a filePaste a linkScreen recordingVoice recordingImport an .srt or .vttSpeaker labelsCustom glossaryFind and replaceTimeline with soundwavesAI summariesChapter markersAsk the videoSRT · VTT exportDOCX · XLSX exportTXT · Markdown exportOne-ZIP downloadShare link, no sign-inWorkspace rolesSwitch project type
The detail lives in the product — a free workspace takes about a minute to set up.
See it in the productQuestions people askabout Gemini 3.5 Transcribe and this tool
Subanana doesn't lock to any single vendor. Every file is routed across continuously benchmarked frontier speech engines — the same class as the models in the headlines — picking the best performer for its source language, and hallucination checks re-run suspect segments on a different engine. As new models ship, the benchmark decides, not the branding.
Per Google's launch post, the model returns transcription data — text with word-level timestamps and speaker attribution — through the Gemini API. Turning that into an SRT file means writing cue-segmentation and formatting code. Here, it's the export button.
Get an API key in Google AI Studio, send pre-recorded audio to the model through the Interactions API — it auto-detects 85+ languages and attributes up to 3 speakers, per Google — then convert the returned words and timestamps into subtitle cues in your own code. The right path if you're building transcription into a product of your own.
Per the launch post: automatic detection of 85+ languages including regional accents, word-level timestamps with speaker attribution for pre-recorded audio, support for up to 3 speakers with more marked experimental, real-time streaming through the Live API, and a Google-reported average word error rate of 2.6% for pre-recorded audio and 4.0% for streaming.
Overall accuracy reaches 98%. Every chunk passes two hallucination checks — silence stays silent instead of becoming an invented sentence — and an optional proofread pass proposes misheard and same-sounding words for you to approve, without moving a single timestamp.
SRT, VTT, TXT, DOCX, XLSX and Markdown — single-language or bilingual, or one ZIP with everything. You can also export the video itself with the captions burned in, up to 4K.
A free account previews the first 15 minutes of each file, 3 files a month, no credit card. Downloading the subtitle file is a paid feature — you check the quality on your own video before paying anything.
Questions people askabout Gemini 3.5 Transcribe and this tool
Subanana doesn't lock to any single vendor. Every file is routed across continuously benchmarked frontier speech engines — the same class as the models in the headlines — picking the best performer for its source language, and hallucination checks re-run suspect segments on a different engine. As new models ship, the benchmark decides, not the branding.
Per Google's launch post, the model returns transcription data — text with word-level timestamps and speaker attribution — through the Gemini API. Turning that into an SRT file means writing cue-segmentation and formatting code. Here, it's the export button.
Get an API key in Google AI Studio, send pre-recorded audio to the model through the Interactions API — it auto-detects 85+ languages and attributes up to 3 speakers, per Google — then convert the returned words and timestamps into subtitle cues in your own code. The right path if you're building transcription into a product of your own.
Per the launch post: automatic detection of 85+ languages including regional accents, word-level timestamps with speaker attribution for pre-recorded audio, support for up to 3 speakers with more marked experimental, real-time streaming through the Live API, and a Google-reported average word error rate of 2.6% for pre-recorded audio and 4.0% for streaming.
Overall accuracy reaches 98%. Every chunk passes two hallucination checks — silence stays silent instead of becoming an invented sentence — and an optional proofread pass proposes misheard and same-sounding words for you to approve, without moving a single timestamp.
SRT, VTT, TXT, DOCX, XLSX and Markdown — single-language or bilingual, or one ZIP with everything. You can also export the video itself with the captions burned in, up to 4K.
A free account previews the first 15 minutes of each file, 3 files a month, no credit card. Downloading the subtitle file is a paid feature — you check the quality on your own video before paying anything.