Hi, we're Team Carat. 🤗 Today, we're looking at an AI short film built with Runway Agent, a game that generates worlds as you play, an AI tool tracing clues in ancient inscriptions, and lip sync that turns one video into three languages.

Runway unveiled Flicker, an AI short film about a limping lamp searching for its place in the world. The official post has drawn roughly 72,000 impressions. In the teaser, a lamp with a broken leg wakes up on a clearance shelf, wanders through a furniture store, and comes face-to-face with a child pushing a shopping cart.

In Runway's production walkthrough, the creator asks Agent for a story about a small lamp struggling through a furniture store. When Agent asks for the emotional register, the creator chooses 'bittersweet and quiet,' then develops scene references, camera movement, the lamp's performance, and dialogue step by step.

PixVerse unveiled PixVerse Game, a system designed to generate worlds while you play. Its official post has reached about 70,000 impressions. The video shows settings shifting between medieval fantasy, a cyber city, and outer space as prompts and choices generate new characters and quests.

PixVerse is proposing an alternative to games where every room, character, and ending is fixed in advance. The footage released so far is closer to a concept trailer illustrating the product direction than raw gameplay, and the company is currently recruiting early users.

Google DeepMind unveiled Predicting the Past Skill for Antigravity. It uses Aeneas, a model for ancient inscriptions, to restore missing text and estimate when and where an inscription was written.

In the demo, entering a Latin curse inscription produces a likely date, region, supporting evidence, and citations. Other examples map the spread of an ancient cult, visualize networks of visitors to an oracle, and trace clues from a ring theft roughly 1,800 years ago.

fal released VEED's Lip Sync 2.0. In the official demo, the same person's video is synced to English, German, and Spanish audio in sequence. Her mouth movements and expressions shift to match the rhythm of each new voice track.

According to fal, the model handles videos up to 10 minutes at 4K without training or fine-tuning the person first. It also includes a sync-strength control for adjusting how far the mouth movements can deviate from the source. For now, these results are based on the maker's own demo.

That's all for today. We covered an agent that helps shape a film, real-time generative games, AI for ancient-history research, and multilingual lip sync. We'll be back tomorrow with more useful AI updates. 😄