Two AI layers, not one

Captions that are already right when you open the timeline.

Speech recognition gets the words. Everything else on the market then chops them every three words and calls it a caption. Insiab adds a second pass that reads the sentence and breaks where the meaning breaks.

No card. Cancel by not coming back.

The difference is where the line breaks

Same audio, same transcript, same three-word setting. One counts words. The other notices that a question ended and a new sentence began.

Splitting every three words

  1. 1ازاي تتعلم
  2. 2المونتاج؟ انت
  3. 3محتاج تفهم
  4. 4الفيديو

Caption 2 ends one sentence and starts the next in the same breath. The viewer reads a question mark mid-line and has to re-parse.

Insiab

  1. 1ازاي تتعلم المونتاج؟
  2. 2انت محتاج
  3. 3تفهم الفيديو

The question stays whole even though it is four words over the target. A shorter caption is better than an unnatural one.

Letting a model near your captions, safely

The obvious objection to putting a language model in this pipeline is that it might invent, drop or reorder something. Here is why it cannot.

Timings cannot drift

The model regroups word ids and never emits a timestamp. Every start and end is computed from the audio, so a hallucinated number is not a failure mode that exists.

No word is lost

The model's answer is treated as a set of cut points, not as the caption list. Words are then taken from the transcript in order, so duplication, omission and reordering are impossible by construction.

One workspace, two surfaces

The dashboard and the Premiere plugin share the same jobs, minutes and settings. Start a job in one and finish it in the other.

What is in the box

Available now

Captions

Upload audio, get a caption track that breaks on meaning rather than on a word counter. Export SRT, VTT, styled ASS, or read it from the API.

In the plugin

Silence cutting

Ripple out dead air across a sequence. Runs in Premiere because it edits the timeline, not a file.

In the plugin

Repeat removal

Find the takes where a line was said three times and keep the cleanest one. Every cut is a normal Premiere edit you can undo.

Coming

AI video

Generate B-roll from a prompt into the same workspace, with the same job history and metering.

Pricing

Billed on audio minutes, not on seats or exports. Minutes are counted when a job finishes, so a failed job costs nothing.

Free

$0/month

Enough to judge the output on your own footage.

  • 30 minutes of audio each month
  • AI caption segmentation
  • SRT, VTT and plain text export
  • One API key for the Premiere plugin
Get started

Creator

Popular

$19/month

For a single editor shipping regularly.

  • 600 minutes of audio each month
  • Every export format, including styled ASS
  • Silence cutting and repeat removal in Premiere
  • Custom caption instructions and presets
  • Priority processing queue
Get started

Studio

$79/month

For teams sharing a workspace and a backlog.

  • 3,000 minutes of audio each month
  • Five seats on one workspace
  • Unlimited API keys
  • AI video generation
  • Usage reporting per member
Get started

Enterprise

Custom

Volume commitments, invoicing and support terms.

  • Custom minute allowance
  • Unlimited seats
  • SSO and audit export
  • Dedicated support channel
Talk to us

Try it on the clip you are arguing with right now.

Thirty minutes a month, free, forever. Long enough to know whether the second layer earns its place.

Create a workspace