- AI Periscope
- Posts
- Apple releases ‘MGIE’, revolutionary AI model for image editing
Apple releases ‘MGIE’, revolutionary AI model for image editing
And: Microsoft delivering Copilot to everyone | Guy used ChatGPT to talk to 5,000 women on Tinder
Exploring below the surface of AI headlines.
Summaries | Insights | Points of Views
In Today’s Edition
Apple MGIE
Image source: VentureBeat | Midjourney
Summary - Apple has launched a demo of "MGIE," a new AI tool that changes how we edit images using simple text instructions. This open-source AI model uses advanced technology to understand and follow user commands, making it easy to tweak photos at a detailed level. Developed with the University of California, Santa Barbara, MGIE shows off Apple's latest AI advancements. This is still in the experimental stage so don’t look for MGIE in an Apple product just yet. Instead, you can read the research paper here, and play with the demo here.
Buoy points:
What is MGIE?: MLLM-Guided Image Editing, a multimodal large language model designed to edit images based on natural language instructions.
Ease of Use: Traditionally, image editing to a professional level required a fair amount of technical skill to use the various software applications on the market. MGIE changes this by allowing users to edit images through simple, natural language instructions.
Versatile Editing Capabilities: The model supports a wide array of edits, from basic color adjustments and Photoshop-like modifications to comprehensive photo optimizations and targeted local edits.
Empowering Creativity: MGIE opens up new avenues for creative expression, enabling users to easily convey their visual ideas in natural language. This lowers the barrier to image editing.
Open Source Accessibility: As an open-source project, MGIE is readily available on GitHub with detailed documentation and demos.
Image editing vs generating: You’ve likely tried image generation tools like DALL-E and Midjourney. The breakthrough that MGIE represents is that it enables the editing existing images with textual prompts.
Image source: Apple | UC Santa Barbara study
POV - Nearly everyone has edited images after they take a picture on their phone. Now image if you could manipulate very localized sections of the image (not just apply saturation or filters to the entire image) using natural language that you type. Going further, integration with speech-to-text capability would enable image editing just by you talking to your phone. However, in my limited use of the demo here, it didn’t work very well, and there were hundreds of requests queued up in front of mine - so be patient.
Microsoft Copilot
Image source: Microsoft
Summary - Microsoft announced today new updates to its Copilot platform, now enhanced with a major update that simplifies user interaction and introduces new capabilities in advanced image creation and editing. Microsoft also introduced the Deucalion model being shipped, although documentation is slim and the model wasn’t even mentioned in their blog announcement. Deucalion seems to be a fine-tuned model designed for richer and faster user experience.
Buoy points:
Copilot's New Look and Feel: A streamlined interface on copilot.microsoft.com and its mobile app makes it easier for users to bring ideas to life and understand the world better.
Explosive Growth in AI Engagement: Over the past year, Copilot has seen over 5 billion chats and images created, demonstrating widespread adoption and sustained growth in Microsoft's Edge and Bing.
Advanced Image Creation and Editing: The new Designer feature in Copilot allows users to customize images with inline editing, offering free enhancements and, for Pro subscribers, additional flexibility in image resizing and regeneration.
Global Accessibility and Utility: Copilot's AI image generation features are now available in English across several countries, and Microsoft is “democratizing our breakthroughs in AI to help make the promise of AI real for everyone.”
Oh, and as a sign of bringing AI to everyone, Microsoft pre-released their Super Bowl ad, to be seen by a worldwide audience on Sunday:
POV - I find myself thinking “what a time to be alive” a lot lately. The pace of advancements in AI that actually make it to your fingertips these days is astounding. I’ve used Copilot a fair bit, but I’m looking forward to diving deeper into their updates.
ChatGPT Romance
Image source: Gizmodo | A. Zhadan
Time to take a break from tech titan updates and share a lighter story with you.
Summary - Aleksandr Zhadan used OpenAI's GPT-2 (and subsequent versions) to create a dating assistant that interacted with over 5,000 women on Tinder. And it worked. He met his wife, Karina Vyalshakaeva in doing so. By programming ChatGPT to match his preferences and handle initial conversations, Zhadan screened potential matches effectively, eventually narrowing them down to his bride. Despite the unconventional start, the couple developed a genuine connection, underscoring the potential of AI to transform personal relationships and dating.
Buoy points:
Innovative Use of AI: Zhadan's application of ChatGPT shows an innovative use of AI in personal relationships.
Customization and Personal Touch: Despite the AI's involvement, Zhadan's effort to personalize the chatbot's responses played a crucial role in building a genuine connection.
Potential Ethical Considerations: This story raises questions about the ethics of using AI in dating, the potential for misunderstandings, and the implications for authenticity in initial interactions. In fact, the article noted that Zhadan wasn’t even reading many of the initial message exchanges with his potential Tinder matches.
AI's Role in Enhancing Human Relationships: The successful outcome of this couple’s relationship suggests that when used thoughtfully, AI can even help you become more efficient… at dating.
She Wasn’t Mad: Karina did not know about the AI-enhanced dating method until several months into the relationship. She wasn’t mad.
POV - Good ending to a unique start of a relationship. Strong relationships are built on trust, and I personally would not endorse this approach. Good on them, though. Would you use this method? What if you were on the receiving side of this?