Belvedir · Launch Video Breakdown: Hook, Pacing & Motion Design

Scene-by-scene timeline & spoken transcript

1. 0:00 – 0:09

   The Hook

   The Private AI Prediction

   “In three years, everyone is going to have a private AI model running on their own compute. They're cheaper, perform better for your custom tasks, and allow you to own your intelligence.”

   On screen

   Zachary Yu Founder, Belvedir

   Camera

   Static front-facing medium shot cutting to medium close-up.

   Motion

   Clean line icons popping sequentially to highlight key value pillars.

2. 0:10 – 0:23

   Problem Agitation

   Current Solutions Are Broken

   “But how do you actually make a private AI model? How do you organize the data to train one? How do you host and route one? The current solutions either deploy expensive FDEs, are too complicated to use, or don't cover the full stack.”

   On screen

   $$$ Expensive FDEs COMPUTE ✓ DATASET CREATION ✕ TRAINING ✓ INFERENCE ✕

   Camera

   Alternating camera angles to stark white graphic cards.

   Motion

   Complex architecture diagram morphing into clear status pills.

3. 0:24 – 0:27

   Product Reveal

   Introducing Belvedir

   “This is why we're making Belvedir. The easiest way to make private AI models.”

   On screen

   Belvedir. The easiest way to make private AI models.

   Camera

   Static frontal shot with centered typography overlay.

   Motion

   Soft fade-in of brand mark and typography.

4. 0:28 – 0:52

   Feature Teaser

   Autonomous End-to-End Pipeline

   “Install our SDK with just a few clicks, and we can collect your traces from your agent, and convert them into task-specific training sets and RL environments. Belvedir then autonomously trains custom models and harnesses, tests them with benchmarks, and privately hosts them, either locally or in the cloud. We then make custom routers between your models, and recursively improve everything as we collect more data and traces.”

   On screen

   > npm install belvedir TRAINING SETS RL ENVIRONMENTS CUSTOM MODELS HARNESSES BENCHMARKS ✓ LOCALLY CLOUD INPUT -> ROUTER -> Fable 5 / Your model / Qwen 3 - 8B Trace Collection -> Datasets + Environments -> Autonomous training -> Hosted inference -> Product Usage

   Camera

   Full-screen clean vector motion graphics.

   Motion

   Step-by-step schematic animations showing trace collection, training, and dynamic model routing.

5. 0:53 – 1:03

   Feature Teaser

   Customer Results

   “We are live with ten companies, and our customers are already seeing up to 2x improvement on their benchmark scores and 10x reduction in their inference costs. All done with less than five minutes of setup.”

   On screen

   2x Improvement On Benchmarks 10x Lower Inference Costs < 5 min Of Setup

   Camera

   Medium frontal cut to profile close-up.

   Motion

   Minimalist typographic metric counters appearing beside speaker.

6. 1:04 – 1:25

   Call to Action

   Call to Action & Outro

   “Private AI models are going to win, but we need billions of them for everyone to have one. Belvedir is the all-in-one platform making the creation of private AI models cheap and easy. Belvedir.ai. Make models like it's nothing.”

   On screen

   belvedir.ai Make models like it's nothing. Belvedir

   Camera

   Camera cut to search bar overlay, transitioning to logo freeze frame.

   Motion

   Animated search box with simulated text input followed by blurred background dissolve.