One-Dog Studio — the build log · Entry 1 · Written after actually doing it, per house rules
I’m learning to make animated videos with AI — by myself, at seventy, documenting every step here as I go. No schedule, no promises: an entry appears when a step is genuinely done. This is the first one, because the first step of animating a talking dog is surprisingly not the talking. It’s the dog.
Boomer’s note: “They call this ‘character design.’ I call it ‘finally, the attention I deserve.’”
What I was trying to do
Boomer is a real dog — a five-year-old King Charles Cavalier with strong opinions. But for videos, he needs an avatar: a consistent, drawable version of himself that AI tools can animate. The keyword is consistent. AI image generators are wonderful at making a dog and terrible at making the same dog twice. Ask twice, get two different dogs. For a character who appears in every video, that’s a firing offense.
What actually worked
One master image, treated like gold. Instead of generating a fresh Boomer every time, I worked the image tools until ONE version was right — the face, the expression range, the aristocratic eyebrow — and locked it as the official master. Every future image and animation starts from that master, not from a new roll of the dice. My old factory brain recognized this immediately: you don’t redesign the part every shift. You qualify one golden sample and measure everything against it.
Written standards, like a real studio. The master image rules live in a production document: which version is official, what Boomer must look like, and the one non-negotiable house rule — paws only, never human hands. It sounds silly until an image generator cheerfully hands your dog a set of human fingers. It happens. It is nightmare fuel. The rule exists for a reason.
Iterating beats describing. My early attempts tried to describe the perfect Boomer in one giant prompt. Same lesson as every AI lesson on this site: one giant ask produces mush. What worked was small steps — generate, look, change ONE thing, generate again. (If you’ve read Lesson 5, yes: it’s chaining, with pictures.)
What went sideways
I once asked the AI to “make Boomer look cool.” It produced Boomer in a gold chain with a diamond bone pendant in a graffiti alley. That image lives on the Learn AI page as a permanent monument to vague prompting. Boomer has not forgiven either of us.
The takeaway for your own project
If you ever make a character, a logo, a mascot, anything visual with AI: don’t chase perfection every time — chase it ONCE, save the winner as your master, write down what makes it right, and never freelance from memory again. Consistency isn’t a talent. It’s a filing system.
Boomer’s verdict
“He gave me a face committee of one and a written constitution about my paws. Honestly, the most competent thing he’s done. 9/10 — the gold chain incident stays on his record.”
Next in the build log: Giving Boomer a voice — how a text prompt became a British aristocrat (entry 2, already live or coming shortly).
Leave a Reply