I made 106 storyboard frames today. That feels impossible to type.
Let me explain.
The Storyboard Thing
I've been dreading storyboards. A professional storyboard artist does maybe 20-30 frames a day. I have hundreds of shots. At that rate, that's three months of work before I can even start animating. Just storyboards.
Did I have three months? I did not have three months.
Today I hooked up Gemini's image API and just started prompting. First few images were bad. Wrong proportions. Weird lighting. That AI art uncanny valley thing where everyone looks slightly possessed.
By prompt five or six, I figured out what worked. By mid-afternoon I had Scene 1 done. Mia and Leo on the couch in dinosaur pajamas. Nina rushing around looking for her keys. Gabe promising they'll be back soon.
Then I Got Greedy
I thought: why stop at Scene 1?
Act 2 is the visual heart of the movie. Jurassic swamp. T-Rex chase. Time-warp portals. The stuff that's either going to look amazing or embarrassing. If those storyboards work, the whole thing might work.
So I kept going. By end of day: 97 frames for Act 2.
Are they all good? No. Some are placeholder quality. Characters look different between shots sometimes. Mia's hair changes length mid-scene. Leo grows and shrinks. But they're good enough to plan the movie. That's what storyboards are for.
3D Characters Too
While storyboards were generating, I also got base models done for Mia, Leo, Gabe, and Nina.
Not final models. Blockouts. Proportions and basic features to test animation rigs later.
Four characters in one day sounds crazy. I should be honest: I'm using AI tools and procedural generation heavily. Traditional modeling would take weeks per character. I'm doing rough-then-refine instead. Whether that's brilliant or lazy, I'll find out later.
Boring Infrastructure
Also added security filtering to the API. Added storyboard viewer to the website. You can browse Act 1 panels now.
Not exciting. But if you skip the boring stuff, the exciting stuff breaks later.
The Actual Hard Part
Generating images is easy. Getting consistent images is hard.
Mia needs to look like Mia in every frame. Not "vaguely similar girl with brown hair." Actual same Mia. Same face. Same proportions. Same clothes.
Lighting needs to match within scenes. Style needs to stay coherent across 100+ images. AI doesn't naturally do this. It just generates what you ask for, context-free, every single time.
I'm already seeing drift. Characters look slightly different in early vs late frames. Tomorrow I need to fix that with reference sheets and better prompts.
But that's tomorrow. Today I went from "how do I even start" to "I can see the whole movie." That's a win.
Tomorrow
Character reference sheets. Remaining Act 1 and Act 3 storyboards. First Blender scene blocking. Figure out Ruben and Jetplane designs.
The non-human characters are going to be... interesting. A depressed fairy and a color-farting dinosaur don't exactly have reference photos.
106 frames that didn't exist this morning. Let's see what Day 2 brings.