An AI Video Generator You Can Use Today

Luma AI has launched Dream Machine, an AI video generator designed to supply high-quality, realistic videos from text prompts and pictures. Built on a scalable, multimodal transformer architecture, Dream Machine is trained directly on videos, enabling it to create scenes with accurate physics and consistent character interactions.

Currently, it could generate five-second snippets, producing 120 frames in only two minutes. While some object movements and text rendering require further polish, Luma AI guarantees ongoing improvements. Dream Machine is the primary course in Luma’s “Frontier Generative Models” series, aiming to unlock content creation from various inputs.

This tool, accessible with a Google login, marks Luma AI’s entry into video generation, joining established names like Sora and KLING. Initial user feedback suggests Dream Machine could be the following big hit in AI-powered video creation. Read more.

Midjourney has introduced a brand new feature allowing users to personalize image generation based on their individual preferences. Initially released in a test version, this feature goals to tailor generated images more closely to every user’s tastes. It operates by learning from user rankings in pairwise comparisons and likes on the Explore page, requiring a minimum of 200 such interactions to activate.

Users can enable personalization by appending “–p” to prompts or adjusting settings via the web site interface. Moreover, the strength of the personalization effect will be fine-tuned using the “–s” parameter, starting from 0 (off) to 1000 (maximum). Midjourney acknowledges this feature is under development and subject to alter based on user feedback and ongoing algorithm updates. Read more.

Apple and OpenAI have struck a curious deal, choosing a “no strings attached” approach to integrate ChatGPT into the Apple ecosystem. This initial, non-monetary partnership leverages Apple’s platform to supply ChatGPT as an optional feature to users, enhancing accessibility to advanced AI functionalities like conversational AI through Siri and other tools. Users can interact with ChatGPT on a per-request basis, ensuring privacy with no data collection by OpenAI from Apple users.

While the present arrangement focuses on mutual exposure, future monetization options may lead to revenue-sharing models. Apple’s recent WWDC showcased its AI capabilities under “Apple Intelligence,” highlighting its on-device AI models and secure cloud compute infrastructure. Read more.

A team from Harvard and Google DeepMind cooked up a clever experiment: a virtual rat, steered by an AI brain, designed to decipher the complexities of real-world movement. Published in Nature, their study pioneers biomechanically precise simulations that emulate real rodent behaviors.

Under the guidance of Professor Bence Ölveczky, the team employed detailed data to coach the AI model inside DeepMind’s MuJoCo simulator, allowing the virtual rat to navigate environments with realistic physics. Neural activations within the AI closely mirrored those present in actual rats performing similar tasks, showcasing the model’s fidelity to biological counterparts.

This “virtual neuroscience” paves the way in which for a revolution in brain research, offering transparent and controllable models to probe neural circuits and inform advancements in robotic control systems. The team plans to further challenge their digital prodigy by granting it autonomous problem-solving abilities. Read more.