Skip to main content
Made with ❤️ by Pixit
Made with ❤️ by Pixit

AI News #79

AI News #79

Pix
Jul 15, 2024 9:46:42 AM

LivePortrait Animates Images with High-Precision Video-Driven Techniques

liveportrait

Story: Researchers from the University of Science and Technology of China have introduced LivePortrait, a novel approach to animating portrait images using video-driven techniques. This method achieves high-precision control and fast inference speed, enabling efficient and realistic portrait animation.

Key Findings:

  • Stitching-based Animation: LivePortrait uses a stitching-based approach to animate portrait images, combining the best-matching patches from a driving video to create smooth and coherent animations.

  • Retargeting Control: The method incorporates retargeting control, allowing users to specify the desired motion and expressions for the animated portrait, enhancing flexibility and customization.

  • Fast Inference Speed: LivePortrait achieves real-time inference speed, processing up to 30 frames per second on a single GPU, making it suitable for interactive applications.

  • High-Quality Results: The animated portraits generated by LivePortrait exhibit high fidelity to the input image and the driving video, preserving the identity and style of the subject.

  • Diverse Applications: This technique has potential applications in various fields, including virtual avatars, video conferencing, and entertainment.

Pixit‘s Two Cents: By leveraging video-driven techniques and retargeting control, this approach opens up new possibilities for creating engaging and personalized animated content. As the demand for virtual avatars and immersive experiences grows, LivePortrait could become a valuable tool for developers and content creators seeking to create lifelike animations efficiently. On our RTX4090 we could create stunning avatars in milliseconds!!

Odyssey Pioneers Hollywood-Grade Visual AI for Immersive Storytelling

oddysey

Story: Odyssey, a startup founded by former Cruise CEO Oliver Cameron, is developing a groundbreaking 'Hollywood-grade' visual AI system to empower professional storytellers with advanced tools for creating high-quality, immersive visual content and experiences. The company aims to revolutionize the way AI is used in storytelling, rejecting the notion of replacing human creativity with algorithms optimized for clicks.

Key Findings:

  • Pioneering Hollywood-Grade Visual AI: Odyssey is at the forefront of developing visual AI that enables the generation and precise direction of stunning video content, including beautiful scenery, characters, lighting, and motion.

  • Empowering Professional Storytellers: Instead of replacing humans, Odyssey's visual AI is designed to be placed in the hands of professional storytellers, providing them with game-changing efficiencies and creativity while maintaining full control over every element in the scene.

  • Experienced Team: Odyssey's team comprises AI researchers, computer graphics experts, and Hollywood artists who have delivered state-of-the-art AI and simulation systems at companies such as Cruise, Waymo, Wayve, Tesla, Microsoft, Meta, and NVIDIA.

  • Rejecting Low-Quality AI Content: Odyssey aims to hold AI to higher standards, rejecting the proliferation of low-quality AI-generated content that lacks the spark and story of human storytelling.

  • Hiring Top Talent: The company is actively hiring for various technical roles, including generative 3D, motion, AI simulation, graphics, 3D reconstruction, ML infrastructure, data engineering, and software engineering.

Pixit‘s Two Cents: By focusing on empowering professional storytellers rather than replacing them, the company is charting a path that leverages the strengths of both human creativity and advanced AI technology. This approach has the potential to turn the the way we create and experience visual content up side down, from movies and television to video games and virtual reality. We are excited how this new venture develops and if it can keep up with their promises.

Microsoft's MInference Slashes LLM Processing Time by 90%

oddysey

Story: Microsoft Research has introduced MInference, a novel approach to prompt inference for long-context large language models (LLMs). This technique significantly reduces processing time by up to 90% while maintaining high-quality outputs, making it easier and more efficient to work with LLMs on resource-constrained devices.

Key Findings:

  • Efficient Prompt Inference: MInference optimizes prompt inference for LLMs by leveraging a combination of techniques, including dynamic sequence length, adaptive attention, and efficient token representations.

  • Reduced Processing Time: By applying MInference, the processing time for prompts with millions of tokens can be reduced by up to 90%, enabling faster and more responsive LLM applications.

  • Maintained Output Quality: Despite the significant reduction in processing time, MInference maintains the quality of the generated outputs, ensuring that the LLMs' performance remains high.

  • Resource-Constrained Devices: MInference makes it possible to run long-context LLMs on resource-constrained devices, such as smartphones and edge devices, expanding the potential applications of these models.

  • Open-Source Availability: Microsoft has made the MInference code and models publicly available on GitHub, encouraging researchers and developers to build upon and extend this work.

Pixit‘s Two Cents: By drastically reducing processing time while maintaining output quality, this technique opens up new possibilities for deploying LLMs in real-world applications, particularly on resource-constrained devices such as phones or other edge devices. As LLMs continue to grow in size and complexity, efficient inference methods like MInference will become increasingly important for ensuring their practicality and accessibility.

Small Bites, Big Stories:
Tags:
Pix
Post by Pix
Jul 15, 2024 9:46:42 AM