This time, my submission isn't a final AI-generated video—it's the process behind it, which I think is equally important. This was my first time using an MCP.

Here’s the setup. I created a standard directory on my computer: 📂 /characters 📂 /environments 📂 /scripts

Inside, I just drop simple Markdown (.md) files describing my characters, sets, and scenes. That’s it.

Using the Melius MCP, my AI assistant reads those local folders and autonomously goes to work. It jumps onto the Melius visual canvas, builds a massive node graph, generates the character "anchors," routes them via Image-to-Image to keep the designs consistent, and spits out full storyboard grids and timed video sequences.

The best part? If I want to add a new character to the movie, I don't even have to touch the UI. I just drop a new text file into the /characters folder, run the command, and the AI automatically expands the node graph and draws them.

My Feedback on the tool: 🟢 The Win: The fact that I orchestrated an entire 6-scene animated short without manually entering a single image prompt. Being able to drive complex visual generation purely from external text data is a massive win for me. 🔴 The Friction: I struggled a bit with the multiple canvases feature. I wanted to create a master canvas that imports generated assets from other canvases, but there doesn't seem to be a simple native way to link them. I had to download the assets to my computer and then manually re-upload them to the new canvas.

https://app.melius.com/projects/f26998a7-d044-4c94-b9fa-706290aef1da/canvas/9905539b-dd1f-4613-8783-ac84a33c6235