Generate media and extract text

Use these tools to generate images and speech, transcribe audio, and extract text with optical character recognition (OCR). Generated files become document assets.

Generate an image

1. Open an editable document while signed in.
2. Click Insert.
3. Click AI image.
4. Enter an image description.
5. Select the dimensions.
6. Click Generate.

The job runs on an eligible Node. When it finishes, AgentsMarkdown stores the result as a document asset and inserts image Markdown at the cursor.

Use prompts that describe the subject, composition, lighting, and intended visual style. Keep dimensions within the limits shown by the tool.

Listen to a document

To synthesize and play the document text, in the toolbar, click Listen. To end playback, click Stop.

Speech runs on your Node. The browser fetches the finished audio through the AgentsMarkdown media route, so the document never needs a relay credential.

To save synthesized speech as an asset through the API, use the speech save endpoint described in the agent API guide.

Transcribe an audio file

1. In Insert, click Transcribe.
2. To include timestamps, select Timestamps.
3. Click Upload audio.
4. Select an audio file.

Batch transcription accepts common browser audio types. For the broadest compatibility, use a mono 16 kHz WAV file.

With timestamps off, AgentsMarkdown inserts plain text. With timestamps on, it inserts a Markdown table with one utterance per row:

Time	Transcript
00:00.000 - 00:03.240	Welcome to the release review.
00:03.240 - 00:07.810	We will start with the build.

Dictate in real time

If the Node supports streaming automatic speech recognition (ASR), Transcribe includes a recording control.

1. Place the cursor where you want to insert the transcript.
2. In Insert, click Transcribe.
3. Click Record microphone.
4. Allow microphone access when the browser requests it.
5. Speak into the microphone.
6. To finish, click Stop.

Segments arrive as the Node recognizes them. Timestamped mode keeps each finalized utterance in its own table row.

Transcription accuracy and delay depend on the model, hardware, and audio quality. Review the inserted text before using it.

Microphone permission

The browser asks before AgentsMarkdown can capture audio. The relay ticket is single-use and tied to the signed-in session.

Run OCR

1. In Insert, click OCR.
2. Click Choose image.
3. Select a PNG, JPEG, GIF, or WebP image.

AgentsMarkdown inserts the extracted text into the document. Review tables, handwriting, and dense layouts before treating the output as final.

Use Node media from MCP

MCP clients can call the same capabilities:

• agentsmarkdown.generate_image
• agentsmarkdown.speak_text
• agentsmarkdown.transcribe_audio
• agentsmarkdown.ocr_image

These tools require an account session token. A share key alone cannot authorize Node jobs. For client setup, see MCP.

Why a control might be missing

If a control is unavailable, check for the following conditions:

• No Node is online
• The Node has not rescanned after a model install
• The required model or runtime pack is missing
• The Node and AgentsMarkdown are signed in to different mere.world accounts
• Your document role cannot edit

Continue with Node troubleshooting.