Tägliche KI-Host-Videoepisoden mit Claude, Gemini und Shotstack generieren

## Make an AI-hosted show: same host, new episode every run A podcast-style vertical episode with an AI host who looks the same every time. Claude writes each episode, Nano Banana 2 puts the host in a new scene, and Gemini Omni Flash makes the host say the lines, lip-synced. Captions come from that speech. No stock footage. Self-hosted n8n only for now: it saves the MP4s on disk. On n8n Cloud, save them to Google Drive instead. ### How it works 1. Your show holds the show, the host, the shots of each episode (one line per 10-second take) and how many episodes to make. 2. An n8n Data Table, made on the first run, logs every episode and keeps the host's picture. 3. With no picture yet, Draw the host makes one, once. Every later episode starts from it. 4. Claude writes the next episodes: a new idea, a title, a scene and one flowing line per take. 5. One Shotstack render per episode: a talking clip per take, back to back, with captions. ### Setup steps 1. Add an Anthropic API key to Claude and Claude (fixes). 2. Add a Shotstack API key to Draw the host, Make the video and Check the render. Free trial: https://shotstack.io/?utm_source=n8n&utm_medium=template&utm_campaign=ai-show&utm_content=workflow_note 3. Make an empty show folder in n8n's files folder (/home/node/.n8n-files in Docker). On an npm install, change that path in Save the episode. 4. Fill in Your show, then click Execute workflow. ### Good to know - Each take costs about 6.25 Shotstack credits, plus 0.5 for the scene and a little for captions and the render: about 20 for a 3-take, 30-second episode. Drawing the host costs under 1, once. - To post every day, set episodes per run to 1 and turn on Every morning.