Skip to content

Media generation

Media generation lets your agent create images and short videos from text prompts, right inside a conversation. Ask for an illustration, a hero image, a social post graphic, or a short video clip, and the agent produces it and shows it inline in the chat.

What you can do with it

  • Images: generate an illustration or graphic from a description, choosing an aspect ratio (square, landscape, portrait, and so on). Images come back inline, right away. The platform offers a curated set of image models; the agent picks the cheapest one that fits your request, or a stronger one when you ask for it.
  • Video: generate a short clip from a prompt, with a choice of aspect ratio and duration. Video takes longer to render, so the chat shows a "generating" state and then displays the finished clip when it's ready. Several well-known video models are available; the agent chooses among them for you.
  • Generated images and videos are saved as files belonging to the agent, with stable links, so they can be reused — for example, referenced in an agent-built app or page — rather than expiring.
  • Everything the agent generates is also listed with the conversation's files, on the canvas, under a short name the agent gives it — ask for a sunset over the lake and the file is called something like sunset-over-the-lake-3f2a1c.png. A picture from three answers ago stays one click away, and you can ask for it again by name: "show me sunset-over-the-lake". Ask for a second version and it is listed as its own file, so nothing overwrites what you already have.
  • When the assistant hands part of the job to a helper or to one of your own agents, the media that agent generates is listed in the same place, marked with the name of the agent that made it — and the assistant can use it too: look at it, put it on a page it builds for you, or open it in the canvas.
  • Generation counts against your usage, and the platform checks your balance before running so a request can't quietly run up a negative balance.

From the phone, iPad, and Mac apps, you can attach photos from your library, take a photo, or attach a document to a message — images ride along with the message and files are uploaded alongside it. Generated images and videos appear as a card you can open full screen, save, or share.

In the browser, click or tap a generated image in the chat to see it full screen. Download image in its top-left corner saves the file.

Availability of image and video generation depends on your plan.

References, variations, and video progress

Attach an image and ask the agent to use it as a reference, or refer to an image it generated earlier. If a reference is missing or unavailable, the request stops with an explanation so the agent can correct it. It will not silently generate a replacement without that reference.

You can request several variations at once. The agent creates the requested outputs, including any preparatory reference images needed for the task. A confirmation step, when shown, changes when generation starts; it does not change the requested number of outputs.

New video requests continue processing even after you close the chat. Videos in the same workspace run one at a time, and the finished files remain available when you return. If one output fails, the agent receives its failure reason and can correct that output without repeating the successful ones.

Some models support only fixed durations or image dimensions. If you do not specify an image format, a supported format is chosen automatically. When a format you explicitly requested is adjusted, the agent tells you which settings were used. The saved result shows the actual duration and aspect ratio.

How the model is chosen

You do not need to know model names. Tell the agent what you want — a vertical clip, ten seconds, with sound, in 1080p — and the platform picks the cheapest enabled model that can do all of it. The agent tells you which model it used in one line.

  • Ask for quality when it matters ("make it cinematic", "the text must be readable", "this is the final version") and the agent picks a stronger model by its description instead of the cheapest one.
  • Ask for a specific model by name and the agent uses exactly that one; if your settings do not fit it, you get an explanation instead of a silent switch.
  • Sound is always an explicit choice — the agent asks when it is unclear whether you want audio.
  • For long, high-resolution or otherwise demanding clips the agent can check first: it names the model it would use and the alternatives, and generates once you confirm.
  • If nothing matches (for example a 30-second 4K clip), the agent explains which requirement no model meets and suggests what to change.

Administrators describe each model's strengths and weaknesses and set its list prices in the admin panel; the description is what the agent reads when it chooses, and the prices only rank the candidates — they are never shown to end users.

How you turn it on

  1. Enable the Image generation and/or Video generation capability on the agent (in the agent's tools). No API keys or provider setup are needed inside the agent — the platform supplies the generation backend.
  2. Just ask the agent in chat ("make me an image of…", "generate a short clip of…"). The agent calls the capability, and the result renders in the conversation.

Media generation is credit-hungry

Image and especially video generations are among the most expensive things an agent can do and can burn through your credits quickly. Keep an eye on your balance and the Usage page, and ask the agent for an estimate before an expensive request.