Images, video and music

Three of the one-turn intents make media instead of text. You arm one, describe what you want, and the result arrives inside the reply and is filed in your Library without you doing anything.

Arming an intent

1. Click Auto ▾ in the composer.
2. Under For this message, pick Image, Video or Music.
3. Describe what you want and send.

Typing /image a heron at dusk does the same in one action, as do /video and /music. The intent lasts for that one message and is cleared by sending. Picking the same one again disarms it.

A row your plan does not cover shows a padlock; clicking it opens the upgrade dialog instead of arming the intent.

What each one needs, and what it makes

Video is the strict one. It is off for Free, Starter and Pro — only Max can generate video. The refusal reads "Video generation requires a Max subscription."

Image generation carries no gate on the intent itself, and image generation is switched on for every plan on this build. Because that is plan configuration rather than code, check your own plan page if the intent ever refuses.

What the intent actually changes

An intent is a posture for one turn, not a mode. It appends a line of direction to the turn and loads a craft skill for that medium, so the model asks itself the right questions about composition, pacing or arrangement. It does not lock the turn to one output type: "make an image and write the caption for it" is a single send, and you get both.

Nor does it pick the model. Which image, video or music model runs comes from your effort setting — see Models and effort, where each effort row lists the Images, Video and Music models it would drive with.

Where the results land

Generated media is inlined into the reply markdown. It does not appear in the deliverables row under the answer — that row is for projects and finished documents from a Build turn.

Every generated image, clip and track is also registered in your Library automatically, so you can find it later without scrolling back through the chat.

An image in a reply has its own actions: Edit image, Mark up and edit (draw on the picture and say what to change), Download, View fullscreen and Open in new tab.

Writing a prompt that works

• Say the subject and the frame, not only the subject. "A heron at dusk, low camera, wide shot, water in the foreground" gives the model something to compose. "A heron" does not.
• Name the medium you are imitating — photograph, ink drawing, technical diagram, 1970s film still. It is the single biggest lever on the result.
• Iterate in follow-ups rather than restarting. Ask for "the same image, warmer light, no bird" in the next message; you keep the thread of what worked.
• For music, describe instrumentation, tempo and mood rather than naming an artist.
• For an edit, use Mark up and edit. Circling the thing you want changed and typing one instruction beats rewriting the whole prompt.

Limits worth knowing

Heads up: you cannot feed a generated video back into a chat message. Add to chat on a video in the Library answers "Video chat attachments are not supported yet." Images and audio re-attach normally.

The other honest caveat: media generation costs credits on top of the turn itself, and it is charged whether or not you like the result. Regenerating three times costs three times. Get the description close before you ask for the expensive version.