How I created an animated webseries pipeline in an afternoon
Sora hooked me fast. What started as a casual experiment turned into an obsession with its creative possibilities. And I’m ashamed to say it, but I bought pro specifically because I wanted to get better at it.
What better excuse for a writer to spend that kind of cash than to write a blog post about it? 🙂
The Idea
For anyone unfamiliar, Sora is a video generation model that turns text prompts—and sometimes images—into short, dynamic clips. It’s essentially a visual storyteller in a box, making it possible for creators to prototype film ideas or sketches in seconds.
It started with a joke commercial script — _Tuck Everlasting’s Arcade._ I needed background characters for the bit, but instead of drawing them by hand, I built a little image-to-image generator that turned my friends’ selfies into childlike cartoons. Suddenly, I had a cast of characters I really connected with.
At the same time, I’d been itching to find a way to make meaningful Sora content within the limits of fifteen seconds. That’s when it hit me: newspaper comics. Tiny, self-contained stories that lived in a single beat. Perfect.
The First Drafts
I opened Obsidian and started sketching ideas. The first generations were rough — uncanny faces, weird lighting, that early‑AI chaos. So I spun up another custom GPT, nicknamed **@Spielburg**, to translate my notes into cinematic vision. Each prompt became a shot list; each scene, a test reel.
Then came the breakthrough: I discovered that uploading an image to Sora meant the video would start from that exact frame. Jackpot. Suddenly I could set the stage myself.
Building the Pipeline
The workflow here moves quickly, so breaking it into steps helps it breathe. First, I created a project in ChatGPT to organize the generated images and threads. Then came a research agent to figure out the best way to assemble and share the clips.
After that, I took some of the kid-sonas my friends had generated and a scenic backplate, using ChatGPT to composite them into a single still. That became the opening frame. From there, the workflow clicked into place: image ➜ composition ➜ motion ➜ cut.
Then, I realized I’d need to share these clips somehow, so I made a research agent figure out the ideal way to put it together.
I took some of the kid‑sonas my friends had generated and a scenic backplate, then had ChatGPT composite them together into a single still. That became the opening frame. From there, the workflow clicked into place: image ➜ composition ➜ motion ➜ cut.
The results speak for themselves:
It was cheap — if I’d paid for every failed generation, the experiment would’ve run about eight dollars total — but the results spoke for themselves. It wasn’t traditional animation, and it wasn’t trying to be. It was a sketch. It was jazz.
The Repeatable System and Future
Once I had that first screen test, I realized I’d built a repeatable creative pipeline. Every step — writing, framing, compositing, generation — was atomic. I could swap pieces, run variations, and reuse the structure again and again.
Custom GPTs turned out to be perfect for this. Each one acted like a specialist: one for script pacing, one for shot composition, one for visual polish. Together, they formed a modular, agentic workflow — a kind of creative assembly line powered by play.
The Future
Now that the bones are in place, I can spin new shorts in an afternoon. Maybe someday I’ll train an agent to handle the whole loop for me. But for now, I’m happy jamming at the console, improvising like a cartoon jazz band.
They’re rough, but they make me grin every time.
The point isn’t that AI made it for me. It was hard work to make these come to life! The fun was in the challenge: to see if curiosity, improvisation, and the right tools could turn a playful idea into something real. It’s that AI let me make it myself — faster, looser, with room to play.
And if that sounds like your kind of chaos, pull up a stool. There’s plenty of wall space left to protest the idea of September.
Leave a Reply