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

Scene-by-scene timeline & spoken transcript

1. 0:00 – 0:22

   The Hook

   Underhyped Gemma 4 12B Announcement

   “Okay so I need to talk about what Google DeepMind just dropped because the headline is underselling it badly. So Gemma 4 12B is a multimodal AI model. Vision, audio, reasoning all in one. Runs locally on 16GB of RAM, which is also a standard laptop in 2026.”

   On screen

   Google Gemma @googlegemma Meet Gemma 4 12B! A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license. Bridging the gap between edge efficiency and advanced reasoning. Here is what's new with Gemma 4 12B: 👇 Gemma 4 12B Unified Transformer 1.4k likes 64 replies

   Camera

   Static front-facing mobile selfie camera

   Motion

   Chroma key green screen overlay over a static tweet screenshot

2. 0:22 – 0:48

   Feature Teaser

   GPQA Reasoning Benchmark Score

   “Now here is the number that completely stopped me. It scores 78.8 on GPQA, which is one of the hardest reasoning benchmarks in AI. For context, GPT-4 level performance was considered the state of art almost I think 18 months ago. And understand that this is a 12 billion parameter model, which is fitting in a device and nearly hitting that bar.”

   On screen

   Google Gemma @googlegemma Meet Gemma 4 12B! A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license. Bridging the gap between edge efficiency and advanced reasoning. Here is what's new with Gemma 4 12B: 👇 Gemma 4 12B Unified Transformer 1.4k likes 64 replies

   Camera

   Static handheld selfie view with hand gestures

   Motion

   Presenter gesturing towards background graphic text

3. 0:48 – 1:17

   Problem Agitation

   Edge AI Market and Enterprise Privacy

   “And I think most of you might not know that the edge AI market is considered to hit 107 billion by 2028. And the reason is exactly this. Enterprises do not want sensitive data leaving their infrastructure. 62% of businesses do not adopt AI because of privacy reasons. So now do you understand? A model this big running fully locally is capable to remove that blocker.”

   On screen

   Google Gemma @googlegemma Meet Gemma 4 12B! A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license. Bridging the gap between edge efficiency and advanced reasoning. Here is what's new with Gemma 4 12B: 👇 Gemma 4 12B Unified Transformer 1.4k likes 64 replies

   Camera

   Static front camera with presenter framing lower-third to mid

   Motion

   Direct-to-camera address emphasizing data privacy pain points

4. 1:17 – 1:47

   Product Reveal

   Architecture Shift and Future of Local AI

   “And of course the architecture is genuinely interesting. So the same model has less complexity, less memory, but it has same output quality. 3 years from now we are going to look back as releases like as the moments AI stopped living AI stopped living in the cloud and started living in the devices. So all in all for me, Gemma 4 12B is huge.”

   On screen

   Google Gemma @googlegemma Meet Gemma 4 12B! A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license. Bridging the gap between edge efficiency and advanced reasoning. Here is what's new with Gemma 4 12B: 👇 Gemma 4 12B Unified Transformer 1.4k likes 64 replies

   Camera

   Static framing with continuous eye contact

   Motion

   Final conclusion takeaway on device-native intelligence transition