Hi Reader,

​
This week, we saw breakthrough models that think in long sequences, talk in real time, and even decode ancient history.​
​

Let’s get into the updates that are defining what’s next.

In today’s newsletter:

• Runway’s Aleph transforms context modeling for video and beyond
• Wan 2.2 launches open-source, local text-to-video for free
• Hedra releases real-time avatars with sub-100ms latency
• DeepMind unveils Aeneas to decode ancient Latin in seconds
• GitHub drops Spark, a coding agent inside your IDE

Let’s dive in.

Runway Aleph: The Future of Context-Aware AI Models

Runway just introduced Aleph, its new foundation model designed to understand and generate sequences across video, text, and more. Unlike traditional models that rely on limited windows or chunks, Aleph is context-native, processing entire timelines in one coherent go.

Key details:

• Context Native Model​
  Instead of a sliding window or chunked attention, Aleph understands entire sequences from start to finish—ideal for storytelling, editing, and scene continuity.

• Built for Video + Beyond​
  Originally optimized for video timelines, but extensible to audio, multimodal tasks, and longform narratives.

• Foundation of Runway Gen-3​
  Powers the next generation of Runway’s Gen-3 video tools used in Hollywood-grade productions.

• Temporal + Spatial Awareness​
  Captures movement, emotion, and visual evolution with higher fidelity than ever.

Try Runway Aleph here👉 https://runwayml.com/​

Wan 2.2: Open-Source Text-to-Video with 1080p Quality

Alibaba has dropped Wan 2.2, a serious new contender in text-to-video and it’s open source, free, and runs locally.

Key details:

• Full HD Generation​
  Outputs videos at 1080p resolution and 30fps, a huge jump from typical open models.

• Run Locally​
  No cloud dependency. Generate cinematic video on your own machine with enough GPU power.

• Open-Source + Free​
  Available to download and build with today—no paywalls or rate limits.

• Great for Tinkerers + Indie Devs​
  Finally, a powerful video model for those who want transparency and local control.

Try Wan 2.2 here👉 https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B​

Hedra Live Avatars: Real-Time AI Characters with Sub-100ms Latency

Hedra has launched Live Avatars, its new streaming model that creates hyper-realistic AI characters who respond instantly, ideal for customer service, live streams, and virtual presenters.

Key details:

• Ultra-Low Latency​
  Response times under 100ms make conversations feel truly live, not pre-rendered.

• Lifelike Expressions + Movement​
  Blinking, talking, smiling, all in sync with audio and intent.

• Use in Video Calls, Twitch, Virtual Hosts​
  Just feed text or voice, and the avatar animates in real time across platforms.

• Built for Scalability​
  Stream hundreds of avatars at once for global deployments in customer service or media.

Try Hedra Live Avatars here 👉 https://www.hedra.com/​

DeepMind Aeneas: AI for Decoding Ancient Inscriptions

Google DeepMind just introduced Aeneas, a specialized AI designed to interpret fragmented Latin inscriptions by drawing context from a vast database of ancient texts.

Key details:

• Historical Contextualization​
  Matches partial fragments to thousands of known inscriptions to reconstruct missing information.

• Trained on Epigraphic Datasets​
  Built in collaboration with historians and archaeologists across Europe.

• First of Its Kind in Humanities AI​
  Opens up new possibilities for ancient language recovery and historical understanding.

• Fast + Accurate​
  Seconds to decode what once took experts days or weeks.

See DeepMind Aeneas here 👉 https://deepmind.google/​

GitHub Spark: AI Agent That Codes Directly in Your IDE

GitHub has introduced Spark, a new autonomous coding agent that lives directly in your development environment and handles complex programming tasks without manual prompting.

Key details:

• Multi-Step Code Execution​
  Spark doesn’t just autocomplete, it plans, edits, and executes multi-line logic flows.

• Built into VS Code and GitHub Copilot​
  Seamlessly integrates with your existing dev tools.

• Understands Context Across Files​
  Reads, modifies, and manages multiple files at once, perfect for large codebases.

• Ideal for Prototyping and Refactoring​
  Can optimize, refactor, or build out entire functions with minimal input.

Try GitHub Spark here 👉https://github.com/​

Google Veo 3: JSON Ad Prompts Going Viral for Brands

Creators are using Google Veo 3 with JSON prompts to build professional-grade ads in minutes, for brands like Nike, Pepsi, IKEA, and more.
​

These aren’t basic animations, they look like real agency-level commercials with cinematic shots, branded storytelling, and camera transitions.

Want the full tutorial and our complete step-by-step workflow?​
Check out the YouTube video we made for you 👇

​

We walk through how to make your own branded ad using Veo3

🎁 Bonus: You can download our exact JSON prompts to use them completely free for your next project.

Get the prompt pack here👉 http://json.sebtips.com​

More wild drops, breakdowns, and tools coming next week.

Catch you next week for another round of breakthroughs.

Stay Creative,

Sebastien Jefferies.

Free: My 100+ AI Toolkit to Supercharge Your Workflow​
​Get your copy here → [Access Now]​