Create a 40-second animated video explaining how Large Language Models (LLMs) work.

Every frame must be generated from code. No stock footage. No video generators. No image editors.

The video must feature cute animated creatures that help explain the process. Make it fun, watchable, and technically accurate.

TECHNICAL REQUIREMENTS
Use Remotion (React/TypeScript) for the animation

Draw all visuals in SVG and Canvas
Use an open-source TTS voice for narration
Synthesize a light, playful score in Python
Export as MP4, 1080x1080 (square for X)
40 seconds total
VIDEO STRUCTURE

0-5s: Hook
Black screen. A small glowing dot appears in the center.

Text types out: "Ever wonder how AI actually thinks?"
The dot expands into a cute, simple creature, like a tiny round blob with big eyes. It blinks and looks around.

5-12s: Tokens (The Input)
The creature receives a sentence: "The cat sat on the..."

The sentence breaks apart into small glowing puzzle pieces, these are tokens.
Each token floats toward the creature, who catches them one by one.
Narration: "First, your words get broken into tokens, small pieces of language."
Cute detail: each token has a tiny face and waves as it flies in.

12-20s: Embeddings (The Map)
The creature holds the tokens and they transform into colorful little gems.

The gems arrange themselves into a 3D map, a colorful cloud of dots.
Similar gems cluster together. "Cat" gems are near "dog" gems. "King" gems are near "queen" gems.
Narration: "Each token becomes a point in a giant map of meaning."
Cute detail: the creature walks around the map, pointing at clusters like a tour guide.

20-28s: Attention (The Thinking)
The creature stops. Multiple tiny creatures appear around it, each one is an "attention head."

They all look at different parts of the map at once.
Lines of light connect the important tokens to each other. "The" connects to "cat." "Sat" connects to "on."
Narration: "The model pays attention to what matters, connecting the dots between words."
Cute detail: the attention creatures nod at each other as they find connections.

28-35s: Prediction (The Output)
The main creature now stands in front of a giant scoreboard.

Hundreds of possible next words appear as floating bubbles: "mat" (biggest), "floor" (small), "moon" (tiny).
The creature reaches up and picks the biggest bubble "mat."
Narration: "Then it predicts the next word and the one after that, and the one after that."
Cute detail: the creature looks proud as it picks the right word.

35-40s: The Kicker
The creature takes the new word and feeds it back into the loop.

The screen fills with a chain of words forming a sentence: "The cat sat on the mat."
The creature smiles, waves at the camera.
Text fades in: "That's how LLMs work. One word at a time."
Narration: "One token at a time. That's the whole trick."
VISUAL STYLE
Dark background (deep navy or charcoal) with soft glowing accents

Cute, simple creature design, round, big eyes, minimal detail, expressive

Tokens and gems have soft glowing edges

Colors: warm gold, soft blue, gentle pink, mint green
Lines and connections drawn in soft neon
Typography: clean sans-serif, white or light gray
Everything should feel warm, playful, and technically clear
AUDIO
TTS voiceover: warm, friendly, slightly curious tone
Script:"Ever wonder how AI actually thinks?"
"First, your words get broken into tokens, small pieces of language."
"Each token becomes a point in a giant map of meaning."
"The model pays attention to what matters, connecting the dots between words."
"Then it predicts the next word, and the one after that, and the one after that."
"One token at a time. That's the whole trick."

Background score: light, playful electronic tune with a soft bounce

Sound effects: soft pops when tokens appear, gentle chimes when connections form

OUTPUT
Export as MP4, 1080x1080
Include the code so it can be re-run with new explanations
Make it easy to add new creatures or scenes later
Build a cute, fast, visually engaging 40-second video that explains LLMs clearly and makes people smile while they learn.