Story: Adobe has announced a new AI-powered ad creation platform that is designed to enhance the productivity of marketing professionals through a user-friendly conversational interface, streamlining access and fostering idea generation within teams. Adobe GenStudio lets marketing teams efficiently devise, produce, manage, and analyze content that aligns with brand identity. Built on generative AI, it empowers any team member to quickly find and generate assets, create variations, and optimize experiences based on real-time content performance insights.
Key Findings:
Creation: GenStudio allows you to quickly generate on-brand content with generative AI.
Content Hub: GenStudio gives marketers a user-friendly content hub that makes it easy to find, edit, reuse, and share assets.
Campaigns: You can use the platform to visualize, plan, and keep track of which campaigns are launching when.
Pixit‘s Two Cents: We love to see that Adobe is publishing yet another generated AI product (other products: Adobe Firefly and Adobe Sensei). Adobe is on a very good way to leverage generative AI to cover its customers’ expectations while ensuring enterprise-grade security and data governance.
Story: Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a new framework that accelerates the process of generating high-quality images using Distribution Matching Distillation (DMD). This technique simplifies the complex, multi-step process of image generation, bringing it down to just a single step. The outcome is a system that not only significantly reduces the time required for image generation but also maintains or even enhances the quality of the visual content.
Key Findings:
Pixit‘s Two Cents: This new framework could help in accelerating image generation. Researchers and practitioners usually need at least 20 up to 80 “steps” to generate images, each step taking multiple seconds. Thus, reducing the number of steps by a factor 30 is significantly reducing the time necessary for generating images. Unfortunately, the team has not yet released the code for the new framework.
Story: Stability AI has announced the release of Stable Audio 2.0, a big update to their audio generation model that brings improved quality, control, and efficiency to the creation of synthetic audio. The new version builds upon the success of its predecessor, offering users a more powerful and versatile tool for generating high-quality audio samples across a wide range of styles and genres. With Stable Audio 2.0, users can now enjoy enhanced control over audio generation parameters, allowing for more precise customization and fine-tuning of the output.
Key Findings:
Improved Audio Quality: Stable Audio 2.0 delivers higher-quality audio samples compared to its predecessor, with more realistic and natural-sounding results across various styles and genres.
Enhanced Control and Customization: The update offers users greater control over audio generation parameters, allowing for more precise customization and fine-tuning of the output to suit specific needs and preferences.
Faster Generation Times: Stable Audio 2.0 features a more efficient architecture that enables faster audio generation times, reducing the waiting period for users and improving overall productivity.
Reduced Computational Requirements: The improved efficiency of Stable Audio 2.0 also translates to reduced computational requirements, making it more accessible to a broader range of users with varying hardware capabilities.
Expanded Creative Possibilities: With its enhanced quality, control, and efficiency, Stable Audio 2.0 opens up new creative possibilities for musicians, sound designers, and content creators, enabling them to explore and experiment with synthetic audio in innovative ways.
Pixit‘s Two Cents: The improved quality and customization options will undoubtedly appeal to musicians, sound designers, and content creators looking to push the boundaries of their creative work. Moreover, the faster generation times and reduced computational requirements make Stable Audio 2.0 more accessible to a wider audience, democratizing the use of synthetic audio technology. As Stability AI continues to innovate and refine its audio generation capabilities, we can expect to see even more exciting developments in the future. Maybe we finally get the well deserved Pixit Jingle?!
DALL-E now lets you edit images in ChatGPT: OpenAI let’s you edit DALL-E images in ChatGPT across web, iOS, and Android.
BBC will stop using AI in Doctor Who promos: The marketing team for the longtime sci-fi series halted its use of generative AI for Doctor Who marketing emails following complaints.
Benchmark Backs AI Video Startup HeyGen at $440 Million Valuation: In the latest funding round, which is not closed yet, HeyGen highlights it’s growth.
Meta’s AI image generator can’t imagine an Asian man with a white woman: On Meta struggling with different races.
Meta’s new AI deepfake playbook: More labels, fewer takedowns: Meta has revised its policies on AI-generated content, aiming for greater transparency rather than content removal. This change involves applying a "Made with AI" label to deepfakes and providing additional context for manipulated content.
Start Using ChatGPT Instantly: OpenAI makes ChatGPT available for instant use without requiring sign-up, aiming to make AI accessible to anyone curious about its capabilities while gradually rolling out the feature and implementing additional content safeguards.
Navigating the Creation of Synthetic Voices: OpenAI discusses the potential benefits and risks of synthetic voices, emphasizing the importance of responsible development and deployment to promote beneficial outcomes while mitigating misuse and harm.
MagicLens: Towards Modular and Composable Multimodal AI: Deepmind introduces MagicLens, a modular and composable multimodal AI system that enables users to interactively explore and manipulate visual data using natural language, offering a flexible and intuitive interface for various applications.
Many-Shot LLM Jailbreaking: Anthropic researchers investigate the susceptibility of language models to adversarial prompts, finding that many-shot jailbreaking can manipulate models to produce harmful or biased outputs, highlighting the need for robust defenses and responsible deployment.
Apple converts your Screen to text: Apple introduces Realm, an on-device AI model that reportedly performs on par with GPT-4 regarding screen context while maintaining user privacy.