Shows a working pipeline where Claude writes shot-level prompts and sends them to a canvas through MCP, with consistency tricks like approving image references before costly video runs.
Key takeaways · AI-distilled
Krea made both films on one Nodes canvas laid out like a traditional production: characters, then locations, then world reference images, then shots. Characters are three-view sheets (front, back, face close-up) and locations are generated empty in the films' 21:9 frame.
Each scene is a Seedance 2.5 node of up to 30 seconds with several cuts. Character and location image nodes plug into it and the prompt refers to them as @Image1, @Image2 in plug-in order, which is how the same kid and classroom stay consistent across shots.
Tomás wrote the prompts in Claude using a prompting agent skillA reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.Full definition → he built and sent them to the canvas through the Krea MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition →. Each shot sets its start and end second, lens in millimeters, camera height and movement, and the actor's expression; one 30-second scene can fill a page.
The team's tips: approve character sheets and locations before spending on costlier video generation, give each character one fixed facial detail such as a mole to spot model drift, and the fortune-teller shots only worked once that prompt part was written in Chinese.
Terms in this piece · Glossary
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.