Image by: Stability AI
Story: Stability AI has introduced Stable Fast 3D, a new generative AI technology that can generate high-quality 3D images from a single image in just half a second. This represents a significant improvement in processing time compared to previous models, such as Stable Video 3D (SV3D), which took up to 10 minutes to generate a 3D asset. Stable Fast 3D accomplishes the same task 1200 times faster.
Key Findings:
Rapid 3D Image Generation: Stable Fast 3D can generate a 3D image from a single input image in just 0.5 seconds, significantly reducing the time and computational resources required for 3D asset creation.
Enhanced Transformer Network: At its core, Stable Fast 3D uses an enhanced transformer network to generate high-resolution triplanes, which are 3D volumetric representations, from the input image. This network efficiently handles larger resolutions without drastically increasing computational complexity, allowing for finer detail capture and reduced aliasing artifacts.
Innovative Material and Illumination Estimation: The model employs an innovative approach to material and illumination estimation, using a novel probabilistic method to predict global metallic and roughness values, resulting in improved image quality and consistency.
Compact, Ready-to-Use 3D Assets: Stable Fast 3D combines multiple elements required for a 3D image, including mesh, textures, and material properties, into a compact, ready-to-use 3D asset, streamlining the asset creation process.
Pixit‘s Two Cents: Since Stable Fast 3D is so much faster than comparable SOTA models and produces better results at the same time, this approach opens the doors to many new applications and much more iterative experimentation than before. We see industries like designers, VR developers and game developers heavily profiting on new possibilities coming from this.
Story: Midjourney has quietly rolled out version 6.1 of its popular AI image generator, surprising the AI community with some super cool upgrades. The update improves image quality, coherence, text, and comes with new upscaling and personalization models. According to this article, “the skin on humans looks more natural” and “if you put words within quotations in a problem it will accurately render those words on the image”.
Key Findings:
Improved Speed: The newest version is 25% faster than v6.0.
Improved Text Capabilities: The newest version is much better in creating accurate text than v6.0 (In a small test, the text in our images was 75% accurate, whereas with version 6, it was only correct in 25% of cases.
Improved Human Realism: Version 6.1 improves the appearance of human figures, making them more lifelike and reducing previous issues with unnatural (artificial) features
Quality Parameter: Version 6.1 comes with a quality parameter (—q), that can be set to 2, to improve image quality (and increase the time to generate the image)
Pixit‘s Two Cents: Midjourney’s v6.1 update is a noteworthy development, especially for those working with human figures in AI-generated imagery (like Pixit). The improvements in realism meet our growing demand for high-quality visuals. At Pixit, we are exicted to see how these advancements can be applied in creating even more realistic employee headshots, fashion photos and much more!
Image: Meta
Story: Meta has introduced AI Studio, a new tool that allows users in the US to create AI versions of themselves or fully custom AI characters on Instagram and the web. With AI Studio, creators and business owners can use these AI profiles to interact with their fans, engage in direct conversations within chat threads, and respond to comments on their behalf.
Key Findings:
Accessible to All Users: Instagram users in the US can start using AI Studio via its website or directly on Instagram by creating a new "AI" profile.
Customizable AI Profiles: Creators can customize their AI based on factors like their Instagram content, topics to avoid, and links they want to share. They can also control features like auto-replies from their AI and specify which accounts the AI is allowed to interact with.
Custom AI Characters: AI Studio supports the creation of entirely new AI characters that can be deployed across Meta's apps, following in the footsteps of startups like Character.AI and Replika.
Pixit‘s Two Cents: For creators and business owners, the AI Studio feature presents an opportunity to boost their limited time and interact with their audience in a more personalized and efficient manner. The ability to customize AI profiles based on specific content, topics, and links allows for targeted interactions. However, we have to see how good the bots actually resemble the users’ tone of interaction and also, at least in terms of celebrities, how well people will take these AI conversations. This definitely takes a lot of authenticity.
Character.AI Cofounders Join Google: Character.AI cofounders Noam and Daniel, along with certain research team members, join Google as part of an agreement that provides Character.AI with increased funding and a non-exclusive license for its current LLM technology, allowing the company to focus on personalized AI products.
Apple Reveals System Prompts for Apple Intelligence: Apple's latest macOS 15.1 Sequoia beta includes system prompts for Apple Intelligence, offering insights into the AI's capabilities, limitations, and ethical considerations, as the company prepares to roll out its AI features across its ecosystem.
Meta Offers Hollywood Stars Millions for AI Voice Projects: Meta is reportedly offering Hollywood celebrities multi-million dollar deals to license their voices for AI projects, aiming to enhance its virtual assistant and audio capabilities while navigating the legal and ethical complexities of AI-generated content.
EU AI Act Architect Says Scope Became Too Broad, Risks Missing Target: The architect of the European Commission's initial AI Act proposal expresses concerns that the legislation's reach has become too broad, potentially missing its intended targets and stifling innovation in the rapidly evolving AI landscape.
OpenAI Releases GPT-4o System Card: OpenAI publishes the system card for its GPT-4o model, providing detailed information on the model's capabilities, limitations, and potential risks, as part of its commitment to transparency and responsible AI development.
Apple Intelligence May Come to EU for Mac, Despite iOS and iPadOS Restrictions: While Apple's AI features may be limited on iOS and iPadOS in the EU due to regulatory concerns, the company is reportedly considering bringing Apple Intelligence to macOS in the region, potentially offering a workaround for users seeking access to advanced AI capabilities.
ByteDance Launches AI Video App: ByteDance, the Chinese tech giant behind TikTok, enters the AI video generation market with the launch of its own app, competing with OpenAI's Sora and other emerging players in the space.
Delays in Nvidia's New 'Blackwell' B200 AI Chip Could Affect Microsoft, Google, and Meta: Nvidia's highly anticipated 'Blackwell' B200 AI chip faces delays, potentially impacting the AI development plans of major tech companies like Microsoft, Google, and Meta, who rely on Nvidia's hardware for their AI initiatives.
Perplexity Unveils Plan to Share Ad Revenue with Outlets Cited by Its AI Chatbot: Perplexity AI announces a revenue-sharing model that will compensate media outlets whose content is cited by its AI chatbot, aiming to address concerns over the use of copyrighted material in AI training and outputs.
Runway Introduces Image-to-Video Generation in Gen3, Transforming Creative Workflows: Runway, a leading AI-powered creative platform, launches image-to-video generation capabilities in its Gen3 update, enabling users to transform static images into dynamic video content.
Nvidia's 'Cosmos' AI Project Mines Vast Video Data for Foundational Model Training: Nvidia's 'Cosmos' AI project aims to scrape and process vast amounts of video data from the internet to train foundational models, raising questions about data privacy, copyright, and the ethical implications of large-scale data mining for AI development.
Apple's New AI Features Reportedly Delayed Until iOS 18.1, Missing iOS 18 Launch: Apple's highly anticipated AI features, including improvements to Siri and the introduction of Apple Intelligence, are reportedly delayed until iOS 18.1, missing the initial iOS 18 launch as the company continues to refine its AI offerings and ensure a smooth rollout.