Skip to main content
A generation is one model run’s output. Generations appear twice: inline in the chat, at the point where the agent submitted them, and in the media rail on the right.

​
In the chat

Runs render as cells in a grid. A batch renders as one grid, laid out like a moodboard. In-flight cells show the model name and a cancel button. Completed cells show the media, a media-type badge, and on hover: A toolbar also appears under a hovered cell. It offers Attach, Quick animate, Upscale, Make a variation, Edit, and Details.
Videos play a preview on hover. Audio renders as a waveform tile. Files that are not media, such as a .zip or .csv, render as a file card with a download button. Retries are grouped in the same cell. Use Previous attempt and Next attempt to compare them.

​
Names and ordinals

Each completed output gets an ordinal in the order the runs landed. Your uploads and video sequences do not take an ordinal. When the agent names a batch, each run also gets a readable name, such as cdg-bag-environments/open-field. The media viewer header shows both. Use the Generations section of the @ palette to refer to an output in your next message.

​
The media rail

The rail shows every tile in the chat, or in the whole project when the scope is Project. Filter by media type or Favorites, and search by prompt, model, or name.
Tile actions, on hover or in the context menu: Cmd+click (Ctrl+click) selects several tiles. An Attach N button then attaches all of them.

​
Cost

The spend chip next to the composer shows the number of generations and the total cost of the chat. Hover it for a breakdown by model or by media type. See Spending caps.

​
Request details

Open a tile’s details to see the model, the endpoint, the request ID, the latency, and the attempt number. Use the request ID when you contact support about a run.

​
Errors

A failed cell shows a short explanation. The agent repairs and retries most failures on its own. See Models and generation for the messages and what they mean.