Describe and transcribe
Turn an image, video or speech into a text note on your board.
Describe writes an AI caption for an image or video. Transcript turns speech in an audio or video file into text. The result lands as a text note in the nearest free space near the source, ready to edit or reuse in a prompt. It never covers other items, so it may sit a little away from the source.
NoteSigned in and paid
Both are paid AI actions. Billing follows the confirmed processing result, so a completed request can be charged even if no usable text is returned. You need to be signed in. They work on media stored on your board, not on GIFs, web links or shared-board media. See Billing.
Describe an image or video
Select the media
Select a JPG, PNG or WebP image, or an MP4, MOV or WebM video.
Choose Describe
Open More actions and click Describe once. Each click starts a separate paid run.
Read the caption
A note reading Processing... appears in free space near the media, then fills with the caption. The run also appears in your generation history.
If it fails, the note is removed and you see an error with Error Details. A request rejected before processing isn't charged. Completed processing can still be charged even when the caption is empty or unusable. While the outcome or billing is unresolved, reserved funds may stay on hold. Check Generation History before starting a new run.
Videos must be fairly short: if the clip is too big you see "Media is too large to prepare safely for Describe." Trim it with Media tools and try again.
Video options
On a video, click the Custom gear next to Describe to open Describe video:
- Model: Gemini 3.8 Flash (default) or Gemini 3.5 Flash Lite (cheaper).
- Processing: Agentic (default) or Static.
- Optional prompt: up to 1000 characters on what to focus on.
- Range start (s) and Range end (s): Static only. Fill in both or neither.
WarningDon't retry an unknown result straight away
If you see "The outcome is unknown. Do not start this request again.", the request may still finish and be charged. Starting it again is a new paid request. If it never resolves, open Error Details, copy the details and contact support.
Transcribe speech
Choose Transcript from More actions on an audio file or a video with sound. Clicking Transcript starts straight away with MAI-Transcribe 2, the default engine. To pick the engine yourself, hover Transcript and choose MAI-Transcribe 2 or Scribe v2 from its submenu. This works the same on audio and video. Each run is a separate paid run. Hover the info icon next to it for an approximate price.
- Files can be up to 3 minutes and 32 MB. For longer recordings, split the file first with Media tools.
- A video without sound can't be transcribed.
- It isn't available on shared boards. Work on a private copy.
The Processing... note is replaced by the transcript, or by "No speech detected".
The caption or transcript is an ordinary text note. Copy it into a prompt, or connect it to a model that takes text input, such as narration.