Start with the words.
Bundled multilingual word lists, context rules and number handling prepare pronunciation locally. Names, links and mixed-language sentences get their own treatment. Uncertain choices are flagged for you to check.
OUR STORY
Your words.
Your rhythm.
Your voice.
A joke can lose everything in one mispronounced word. We’re building crushedwithit to help your material reach the screen with its personality intact.
Hear the thinking behind itOn the App Store for iPhone, iPad, Mac and Apple Watch.
01 / MADE FOR THE WAY YOU SPEAK
Your language.
Even between languages.
English in one breath. Hindi in the next. A city name, a brand, a word only your friends use. Your material should have room for all of it.
- English English
- Hinglish Hindi + English
- Hindi हिन्दी
- Telugu తెలుగు
- Tamil தமிழ்
- Kannada ಕನ್ನಡ
- Marathi मराठी
- Punjabi ਪੰਜਾਬੀ
Written one way.
Said the way you mean.
Roman Hindi can look like English to a video model. “Main” might be read as the English word, even when you mean मैं.
Our pronunciation layer uses language-specific vocabulary and context to prepare a spoken version on your device. Hindi words can travel to the model in Devanagari while English words stay in Latin letters. The same approach uses the appropriate scripts for Telugu, Tamil, Kannada, Marathi and Punjabi.
Your captions keep your original writing. In Script and cuts, “Said as” lets you inspect the pronunciation, check uncertain words and make a correction.
A PRONUNCIATION EXAMPLE
YOU WRITE · CAPTIONS KEEP THIS
Bhai, main client ko bola ki lunch pe milte hain.
SAID AS · PREPARED FOR THE VIDEO MODEL
भाई, मैं client को बोला कि lunch पे मिलते हैं।
Hindi words in Devanagari. English words in Latin letters. One mixed-language line.
Illustrations of pronunciation choices, not an audio preview. Generated speech still needs your review.
These are language choices for writing and reels. Dictation and live speech recognition depend on the language models available on your device; Hinglish uses an Indian English fallback. Rendering quality varies by language and by line.
02 / BETTER INPUTS. MORE FAITHFUL PERFORMANCES.
Machine learning.
With you in the loop.
Speech recognition and video generation do the heavy lifting. Vocabulary, context and your corrections help those models understand what you intended.
Remember your way.
Type how a word should sound, choose an alternative or keep it in English. Remember the correction for one bit or across your material. Your pronunciation list stays available locally and syncs between your devices through your iCloud.
Give the model context.
For reels, the model receives the prepared dialogue and language-specific delivery guidance. For rehearsal and transcription, the recogniser gets hints from your set’s nouns and punch words.
Check, then refine.
We compare the pronunciation layer against reference lines, check that English words survive unchanged, and use rendered samples to examine what the video model actually says. Native-speaker listening matters where automatic transcription is unreliable.
Better results don’t require us to train on your private material.
Personal corrections improve the instructions used for your future reels. They don’t retrain a shared model. We improve our vocabulary, rules and delivery guidance separately, and we don’t train models on your notes, scripts or recordings. Video partners have their own terms, explained before you consent and in our privacy policy.
Our bundled language resources build on AI4Bharat’s Aksharantar and the SCOWL English word lists. Credits and notices are available in the app’s Acknowledgements.
03 / FROM A LINE TO A LONGER PERFORMANCE
Make the reel.
Keep the good parts.
Film yourself with a prompter, or use a Personal Double: a generated likeness of you, made from a recording you approve. Both paths keep your script at the centre.
LONGER PERSONAL DOUBLE VIDEOS
A whole performance.
Made in manageable parts.
Long scripts become a sequence of short renders. Review the script and cuts, check the price, then follow each part as it’s made. The app joins the finished parts into one video, with the individual clips available too.
Reference material and continuity guidance help carry the performer and scene from one part to the next. Check the joins before sharing; generated continuity can vary.
Part 1
First bitPart 2
Next bitPart 3
The closer
First bitPart 2
Next bitPart 3
The closer
One whole video
Illustrative sequence. Length and part limits depend on the rendering service.EDIT WITHOUT STARTING OVER
One part sounds off?
Fix that part.
Remake a selected part, compare takes and choose which one belongs in the full video. The existing take stays until you accept its replacement.
On supported generated parts, correct up to five words in one pronunciation edit. The app quotes the cost first. Where an edit isn’t available, you can remake the part.
DIRECT THE DETAILS
Your script.
Your look. Your say.
Choose your own approved looks and references, inspect the scene preview, and review the exact prompt. Adjust the script, cuts, framing and delivery before committing credits.
Saving a pronunciation changes future instructions. A video already made needs an edit or a new take to say it differently.
FILM YOURSELF
Your camera.
A prompter that listens.
A voice-paced prompter holds when you pause. Record takes, trim by selecting the first and last word, and export with captions and highlighted punch words. Filming, trimming and caption export run on your device.
Takes stay local. A clip you keep can sync to your private iCloud.
SHARE WITH CONTEXT
Made by your double.
Identified as generated.
Generated exports carry signed provenance. The sharing flow checks it before export. If you turn the visible AI label off, the export must pass its embedded Content Credentials check; if embedding fails, the visible label is kept.
Share a whole reel or individual parts, or use a Private Mic link to collect reactions from people you invite.
Generated videos use online services and credits. Check the quote before rendering and review the finished speech and picture before sharing.
Watch Sachin’s set04 / SAFE & SECURE
Your material is personal.
We build around that.
On-device tools. Your private iCloud. Clear choices when a feature needs the cloud.
No app account to create.
You don’t give us an email address or phone number to use the app. Apple handles purchases; your private iCloud connects your devices. A stage name helps personalise your writing.
Useful work stays on device.
Note transcription, rehearsal, room analysis, pronunciation preparation, self-filming, trimming and captions run locally, with the required device capabilities and permissions.
Your library. Your iCloud.
Written material and kept media can sync through your private iCloud. We cannot read that library. Your personal pronunciation choices follow you through Apple’s iCloud sync too.
You choose what leaves.
Writing help sends the relevant text to Anthropic through our server. Personal Double creation, approved references and generated reels use the video service named in the app, currently BytePlus or Higgsfield. You see the consent information before sending.
Protection with purpose.
The service checks app and purchase authorisation. Requests use HTTPS. Generated sharing checks signed provenance. There is no advertising or third-party tracking SDK in the app; optional activity counts start off.
Control beyond “accept”.
Export your work, remove references or your double, and use Delete everything when you want to leave. The app reports deletion failures. Provider copies, shared links and files already sent to others have limits explained in the privacy policy.
The website has a different job.
If you ask us to stay in touch, the website collects the contact details you enter so we can reach you. Public reel likes and moderated comments also store the information needed to work. That is separate from using the app without an email or phone number.
BUILT FOR YOUR NEXT LINE
Keep what makes
you sound like you.
For iPhone, iPad and Mac, with Apple Watch as your stage remote.
One subscription. Different tools for different moments.