Pictures and speech with your provider

Select text and press Ctrl+K, or right-click and choose Edit with Assistant. Beside Shorter, Rewrite, Fix Grammar and Translate are:

Imagine

Makes a picture of the selected words, or of what you type, and puts it after them.

Transcribe

Turns what you say, or an audio file, into written text at that place. The text comes written up, with punctuation and paragraphs; turn off Written Up for the words as spoken.

Describe Picture

With the cursor on a picture, writes its alt text.

Edit Picture

Changes a picture as you ask. The changed picture is saved as a new file and the link points to it; the old file stays.

The note’s right-click menu has them too: Imagine Here or Imagine from Selection, Transcribe Here, and on a picture Describe Picture and Edit Picture. Each shows the change to accept or reject, like any edit. See Edit a selection in place.

• Speak, the microphone: record yourself, or choose an audio file. Your words go into the message box, to send as they are or after a change. When two or more people speak, each gets a label, such as Speaker A.
• Imagine a Picture, the picture button: makes a picture of what you typed, right in the chat. Save to Vault keeps it, and Put in Note shows the change to the open note, to accept. Nothing goes to the chat’s model.

Imagine, at the top of the assistant, keeps every picture made or brought in. Make more there, with a prompt, a shape and one to four at a time, then save them to the vault or put them in a note. Bring In a Picture, or drop one in, to clean it and save it.

Press a picture, in Imagine or in the chat, to see it large, with Put in Note and Save to Vault under it. Esc, Close or a press beside the picture closes it.

Saved pictures go in the Attachments folder at the top of the vault.

Each action uses a provider you added, with your own key. The recommended provider makes pictures, changes them and turns speech into text; its models are in Choose a provider. Transcribe’s write-up and Describe Picture use the chat’s model.

OpenAI, Google Gemini, OpenRouter and Together make pictures too, and they and Groq turn speech into text. Edit Picture works with the recommended provider. Models for Transcribe and Draw, on the gear’s AI Provider page, picks the model for each, and Imagine, the chat box and Edit with Assistant all use those. See Bring your own key.

Every picture has Remove Metadata, on by default. Before a picture is saved, it takes the picture’s metadata out, in your browser. The picture itself doesn’t change. Hover over the switch to see what that covers. Turn it off to keep the picture as it came.