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

AI News #76

AI News #76

Pix
Jun 24, 2024 10:06:55 AM

Runway ML Introduced Gen-3 Alpha: A New Video Generation Tool

runwayml

Story: Runway has introduced Gen-3 Alpha, its latest text-to-video (and image-to-video) model that can produce high-fidelity, controllable videos. The company says that the model delivers a “major” improvement in generation speed and fidelity over previous models (i.e. Gen-2), as well as better controls over style and motion. While the model is able to produce 5 and 10 second long videos, Runway has not shared where the training data is coming from.

Key Findings:

  • Enhanced Video Quality: Gen-3 Alpha provides advancements in video generation quality, offering more realistic and detailed outputs. It is capable of creating highly expressive human characters with a wide range of actions, gestures, and emotions.

  • Multimodal Training: The model is trained jointly on videos and images, enabling it to power Runway's various tools, including Text to Video, Image to Video, and Text to Image, along with existing control modes like Motion Brush and Director Mode. This multimodal training infrastructure supports fine-grained control over video elements.

  • Improved Temporal Control: Gen-3 Alpha features highly descriptive, temporally dense captions, allowing for imaginative transitions and precise key-framing of scene elements. This enables creators to achieve more dynamic and fluid video content.

  • Safety and Provenance Standards: The model includes new safeguards such as an improved in-house visual moderation system and adherence to C2PA provenance standards, ensuring ethical use and copyright compliance of generated content.

  • Unclear Training Data Sources: While the advancements in Gen-3 Alpha are impressive, there is a lack of transparency regarding the origin of the training data (videos and images). This raises concerns about the ethical and legal implications of the data used for training the model.

Pixit‘s Two Cents: Gen-3 Alpha’s release represents a leap forward in AI video generation, providing creators with another powerful tool to produce videos using nothing but AI. The recent interest and development of text-to-video models (among others: OpenAI’s Sora, Google’s Veo, Luma’s Dream Machine, Adobe, and Kuaishou) has let to models getting better and better in generating images and we are looking forward to further improvements.

Dream Machine: Luma’s AI Video Generator is Taking Social Media by Storm

luma ai

Story: After launching the beta of its AI video generation model called Dream Machine, the tool by Luma AI went viral. Dream Machine creates high-quality videos by using text prompts and images. The company promises that users can use the model to create up to 120 frames in 120 seconds.

Key Findings:

  • Advanced Video Generation: Luma’s Dream Machine uses generative AI to convert text prompts into high-fidelity, captivating videos, making it easier for users to produce visually striking content.

  • Customizable Content: Users can personalize their videos by adjusting various parameters, such as style, motion, and elements within the scene, to suit their specific needs and preferences.

  • Integration with Social Media: The tool integrates seamlessly with major social media platforms, allowing users to share their creations instantly and boost their online presence.

  • Security and Ethics: Dream Machine incorporates safety measures to ensure that generated content adheres to ethical standards and avoids inappropriate or harmful material.

Pixit‘s Two Cents: Again, the text-to-video revolution is mind-blowing and it’s funny to see the videos people are creating using the new tool. As reported by Handelsblatt, the tool is used as a “Meme Machine”, creating videos from existing memes. For example, see here.

OpenAI Could Become a For-Profit Business

openai logo

Story: OpenAI, originally established as a non-profit AI research lab, is reportedly considering transitioning to a for-profit business model. CEO Sam Altman has discussed this potential shift with shareholders, exploring the possibility of becoming a for-profit benefit corporation.

Key Findings:

  • Potential Transition: OpenAI is thinking to change its governance structure to become a for-profit benefit corporation. This move would allow the company to pursue commercial opportunities more aggressively, aligning with other AI companies like Anthropic and xAI.

  • IPO Considerations: The transition could pave the way for an initial public offering (IPO), potentially valuing OpenAI at $86 billion. This move would also allow CEO Sam Altman to take a personal stake in the company, which has been a point of interest for investors.

  • Revenue Growth: OpenAI has achieved significant financial milestones, surpassing $2 billion in annualized revenue as of December 2023. This rapid growth is driven by strong demand from business customers utilizing OpenAI's generative AI tools, such as ChatGPT.

  • Regulatory and Lobbying Efforts: To navigate the increasing regulatory landscape, OpenAI is expanding its lobbying department, aiming for a 50-person global affairs team. The company emphasizes that its goal is to ensure AI benefits humanity, rather than merely maximizing profits.

  • Strategic Alliances: OpenAI has formed strategic partnerships, notably with Microsoft, which has invested up to $13 billion in the company. This alliance has led to the integration of OpenAI's models into Microsoft’s AI Copilot for enterprise users of Microsoft 365.

Pixit‘s Two Cents: A potential shift from non-profit to profit model reflects the rapid growth and the expanding commercial potential of AI technologies. While this transition could unlock new resources and opportunities for innovations, it also raises questions about maintaining the company’s original mission of ensuring AGI benefits

Small Bites, Big Stories:
Tags:
Pix
Post by Pix
Jun 24, 2024 10:06:55 AM