Make session social graphics in three sizes

A webhook starts it from a session record in Airtable. The workflow creates a placeholder row in your creatives table, then runs three branches at once, one each for the 1:1, 4:5 and 9:16 sizes.

In each branch an OpenAI model writes an image prompt from a training graphic you chose, the speaker's cutout headshot and name, the session title and the summit dates. The prompt goes to the linked Fal.ai image model, checked every 10 seconds until the image is ready. The three images are merged and attached to the creative row, which is set to To Do. Costs are logged. If a step fails, the session is marked Error.