Documentation

Chat

Work towards an image by conversation, and see why the orchestrator did what it did.

Chat is for when you do not know quite what you want yet. You talk, and an orchestrator model decides whether to answer with text, generate an image, or both.

  • Keep as many separate conversations as you like, each with its own settings.
  • Set the model, temperature and aspect ratio per conversation.
  • Open the orchestration trace to see why it did what it did: the reasoning, the prompt it actually sent, and how long each step took.
  • Download, edit or retry any image without leaving the conversation.

If you have not added a key for an orchestrator, Chat falls back to simple keyword detection to guess when you are asking for an image. It works, but it is noticeably blunter than letting a model decide. See setting up API keys.