🖼️ Ideogram 2.0 arrives to take on Midjourney

Hey AI team! We’re back with a new edition of Five Minute AI this Friday ⏰

On today’s menu:

  • Yet another new AI image generator → Ideogram 2.0 arrives to take on Flux and Midjourney

  • xAI launches Grok 2 in early beta

  • Ride the trend of the moment’s color: neon green! #brat

  • McDonalds makes a video ad entirely with AI, watch it at the end of the newsletter

Ready? Let’s go...

THE MAIN STORY

Microsoft unveils new AI models

On Tuesday, Microsoft launched the Phi-3.5 family of models.

  • Phi-3.5 Mini: This model stands out on multilingual tasks, beating Llama 3.1 8B on several benchmarks.

  • Phi-3.5 MoE: Specialized in reasoning, this model compares favorably with Google’s Gemini 1.5 Flash.

  • Phi-3.5 Vision: Built for multi-image visual tasks, this model rivals GPT-4o in performance.

All the models get a 128,000-token context window, making them well suited to processing long documents and handling complex conversations.

The model has 4.2 billion parameters, giving it significant learning and generalization capacity. A built-in image encoder also lets the model process visual information efficiently.

With these models, we’re getting closer and closer to a future where powerful AI can run locally on our small electronic devices: phones, watches, earbuds…

Ideogram 2.0 arrives to take on Flux and Midjourney

Ideogram just released its v2.0 text-to-image model for free to all users.

Here are the main new features:

  • Color palette selection

  • Better text rendering

  • Search across more than a billion public Ideogram images

  • Better image consistency (human figures, eyes, hands)

Ideogram offers five distinct image styles: General, Realistic, Design, 3D and Anime. Among these, the Realistic style got notable upgrades for more authentic rendering of human textures, while the Design style now delivers much better text rendering.

Ideogram also offers a free tier, letting users generate about 40 images or 10 prompts a day.

You can see some examples of its improved capabilities in action.

xAI launches Grok 2 in early beta

xAI has started rolling out early beta access to Grok 2, a new AI model that uses real-time data from X and Flux.1 to generate AI images.

  • Available to a select group of X premium users

  • Access to real-time information from X

  • Two modes: normal and “fun”

  • Ability to reference specific tweets when gathering information

As you can see in the gif below, the image rendering + animation with Luma Dream Machine is pretty wild… Grok uses Flux Pro to generate the images, and there’s very little filtering on what you’re allowed to generate.

The quality is bad because I had to compress the gif

WHAT THE RESEARCH SAYS

Google unveils HeAR, an AI model that detects diseases from sounds

Google Research has developed HeAR (Health Acoustic Representations), a foundation AI model for bioacoustic analysis:

  • Trained on 300 million anonymized health audio samples

  • Includes a cough-specific model trained on 100 million cough sounds

  • Aims to extract health information from body sounds (coughing, breathing, voice)

  • Available to researchers to build custom models

Potential applications:

  • Early screening for diseases like tuberculosis

  • Monitoring chronic diseases like COPD

  • Remote diagnosis via smartphone

Salcit Technologies, an Indian company, already uses HeAR in its Swaasa app to improve early tuberculosis detection by analyzing cough sounds.

The goal is to make diagnosis more accessible and affordable, especially in regions with limited access to care. This approach is backed by organizations like The StopTB Partnership.

Interested researchers can request access to the HeAR API to explore its potential across various health fields.

LEARN TO MASTER AI

Use the color of the moment #brat

Neon green is the standout color of the moment. This bright, bold shade took off thanks to American singer Charli XCX, who chose it for the cover of her album “Brat,” released at the start of the summer. Since then, the eye-popping shade has become a full-blown fashion phenomenon, embodying the rebellious, energetic spirit of the season.

While the trend is mostly about writing in pixelated letters on a green background (Kamala Harris adopted the concept), the phenomenon has also made this shade the color of the summer, especially in fashion and makeup (face and nails).

To refer to this color in image generators, you need to use the term “lime green“. Let’s try it with Flux Pro; by the way, I made you a tutorial on how to use Flux’s AI models.

Prompt: Ultra-realistic portrait of a Eurasian teenage girl with lime green nails, set in a minimalist and avant-garde scene with studio lighting and subtle lime green neon lights in the background.

THE LATEST TRENDY TOOLS

Palette: A handy little tool for colorizing black-and-white photos. It offers a free trial with no sign-up, and the results are perfectly decent. Perfect for impressing the family at Christmas by recoloring old photos.

I generated this black-and-white image of Paris with Flux Pro (prompt: black and white photography of a photo of Paris eiffel tower before major urban changes)

Once converted in palette, here’s the result:

You can get rid of the watermark using Watermark Remover (though it can leave a small strip). If you want to download the photos in HD, you’ll need the $9/month subscription.

💡 OTHER TOOLS:

  • Kerlig: Use AI in Slack, Figma, Gmail, Outlook, LinkedIn, Code and more on macOS.

  • Heatseeker: Test product concepts and marketing messages before launch with simulated experiments.

  • Aicado: Build your own custom chatbot with private chat history and embed it on any site. Personally, I love the avocado logo 🥑

  • MolyPix: Generate easy-to-edit graphic designs from text. Perfect for making posters, invitations, postcards, birthday cards…

  • LTX Studio Character Animation: Full control over character dialogue and facial motion capture for AI filmmaking.

  • Hermes 3: Open-source AI model optimized for reasoning.

  • Zed: High-performance, collaborative code editor.

  • micro1: AI recruiting engine for finding the best engineers.

  • sleekflow: Automate your customer interactions with custom conversational workflows.

IN THE NEWS

  • WhatsApp is working on usernames as an alternative to sharing phone numbers.

  • McDonald’s Instagram account was hacked and posted a crypto scam, already netting the hacker $700,000.

  • Anthropic’s Claude tops $1 million in revenue on its mobile app.

  • Midjourney has opened its web-based AI image editor and new tools to everyone, with 25 free image generations for new users. If you haven’t had a chance to try it yet, now’s your chance!

  • Groq has released a new API for a streamlined version of OpenAI’s Whisper speech recognition model. It’s 240 times faster and much cheaper, so if you have audio transcription to do, it could be a great deal.

  • A new McAfee tool can now detect AI-generated audio.

  • Nvidia researchers have created an AI that can predict thunderstorms with kilometer-scale precision.

  • Gemini in Gmail can now help you improve your email drafts.

YOUR DAILY DOSE OF CONTENT

1) McDonalds: Discover the new era of advertising. Thanks to recent advances in AI video, it’s now possible to create synthetic ads that meet the standards of the biggest brands. McDonald’s offers a striking example in Japan with an ad generated entirely by AI, likely using Midjourney and Luma for the animation.

Click the image to watch the video

  1. Anthropic just released two new resources to learn how to build prompts for using AI

  2. How hackers create video game cheats using AI

THAT’S ALL FOR TODAY

Thanks for reading the work of our purr-sistent team 😼

💡 Help me improve or come up with new ideas → [email protected] or @mattioo81

👍️ Like what you’re reading? Subscribe!