• NATURAL 20
  • Posts
  • Midjourney's Dynamic 'Vary (Region)' Tool

Midjourney's Dynamic 'Vary (Region)' Tool

Midjourney Unveils 'Vary (Region)' Infill Tool, Redefining Creative Possibilities through Advanced AI-Driven Image Generation

Author

Wes Roth
August 24, 2023

Today:

Midjourney adds new ‘vary (region)’ feature to rival Photoshop Generative Fill

Midjourney is stepping up its game with a new feature called "vary (region)," and it's shaking things up for Adobe Photoshop. How? It lets you change specific parts of a generated image, adding or replacing elements with new stuff, all based on what you type. Yep, it’s AI-powered and it's darn slick.

Adobe's had a monopoly on the editing game for ages with their pricey Photoshop and its shiny new Generative Fill. But Midjourney’s new tool is stealing the spotlight. For just $10 a month, you can get your hands on some similar magic. There's a catch though: you can only use it on images generated by Midjourney itself.

So what's it like? Click the "vary (region)" button in Discord, and you can pick and choose the areas you want to shake up. You can also flip on “remix mode,” type in a new prompt, and boom—you get a whole new look in that area.

Google introduces 11 new security features for Workspace (some AI-powered)

Google's upping its security game, folks. They just dropped the news that Google Workspace is getting 11 fresh security features. Why? Cyberattacks went up by 38% this year and cost a whopping $4.3M per breach. Yikes!

Here's a quick and dirty breakdown of what they're rolling out:

Zero Trust & DLP: Simply put, even if bad guys get in, they can't do much. Google's adding more layers of security beyond just logging in.

AI-powered Labels in Drive: Google Drive's gonna auto-label some of your docs, making them even safer.

Smart DLP in Drive: Security levels adjust based on stuff like where you are and what device you're on.

Better DLP in Gmail: They're being kinda hush-hush on the details, but it's about stopping folks from sharing stuff they shouldn't.

Digital Sovereignty: Fancy term for deciding where your data lives (US, Europe, etc.) and who can touch it.

Cyberthreat Head-On Tackling: Google's gonna make some big accounts use two-step login (because it's way safer) and not let one lone admin have all the power.

AI-boosted Gmail: Google's AI might stop your coworker George from spamming everyone with lunch updates. Maybe.

Easy Logging with Chronicle: Got Workspace? Google's letting you send logs to Chronicle for a closer look.

Google's bringing more security goodness, mainly for the big guys but maybe the smaller players too. No exact dates, but better safe than sorry, right?

Microsoft is “experimenting” with ways to work AI into bedrock Windows

Microsoft's gearing up to add more AI mojo to Windows 11. Come fall, they're dropping "Windows Copilot," but that's just the tip of the iceberg. They're also tinkering with the usual apps like Photos, the Snipping Tool, and even the old-school Paint to pack them with some AI magic.

What's the big deal? Imagine snagging text right from your photos or screenshots using a feature like OCR. That's on the menu. The Photos app might even start picking out faces and stuff, making it a breeze to cut them out from backgrounds.

And MS Paint? It might start whipping up images if you just tell it what you want. Kinda like what those fancy programs like Photoshop do. Heck, they've got something called the Bing Image Creator that's already cooking up images with AI.

Twilio Segment teams with Amazon SageMaker on new customer predictions product

Twilio Segment is stepping up their game. They're teaming up with Amazon SageMaker to launch CustomerAI Predictions, a shiny new tool for marketers. Instead of letting customer data gather dust, this tool will help companies predict how different groups of customers might act. Think of it as a crystal ball to spot who's gonna buy what next.

Segment collects info from website clicks, and Twilio backs it up with communication data. This combo helps companies figure out who's totally into their products and who might need a little nudge. They're leaning on Amazon SageMaker's smarts to build this tool. And word on the street is, they've got an AI email tool coming up. So, marketers could predict customer behavior and then whip up a tailored email based on that data.

CoDeF: Content Deformation Fields for Temporally Consistent Video Processing

Researchers came up with a new method called CoDeF. Think of it like having a main picture (they call it a canonical image) that represents your whole video. Then, there's a guide (temporal deformation field) that shows how this picture changes over time to match each frame in your video.

By working with this main picture, you can apply changes (like filters or object tracking) just once and then easily spread them throughout your video. The cool part? It does this super consistently, even for tricky stuff like water or fog. In tests, it beat out other methods, making video edits look smoother and more natural. Boom! Video magic!

Making LLMs lighter with AutoGPTQ and transformers

Quantization is a promising avenue for reducing the computational demands of large language models without compromising too much on performance. With GPTQ, users can now run large models even on limited hardware, potentially expanding the reach of state-of-the-art NLP solutions to a wider range of devices and platforms.

Researchers are optimistic about the future of quantization in NLP and machine learning as a whole. As models grow larger, it becomes more imperative to find efficient ways to deploy them. The integration of the AutoGPTQ library in Transformers is just the beginning.

Open-Assistant Llama2 70B SFT v10

This is all about a chatbot model that's been fine-tuned for Open-Assistant. They trained it in two main parts. First, they put it through coding tasks and fake instructions. Second, they polished it up using real-life examples they got from folks on their website.

The model's got chops in English and can do a bit in several other languages like German and Spanish. They've also switched up the way you can chat with it to make it easier to use in different apps.

Nous-Hermes-Llama2-70b

The Nous-Hermes-Llama2-70b is a language model that's been souped-up by a team called Nous Research. They worked with a bunch of other folks to make it better than before. This model can talk more, dream up fewer crazy things, and doesn't have OpenAI's usual filters. It's good at a whole bunch of language tasks, like writing stuff or understanding complex orders. They used a fancy computer to make it smarter and threw in a whole bunch of different types of data to train it.

The model did okay to pretty good on a bunch of tests that measure how smart it is. You can use it for different projects, like chatbots or other stuff involving language. They're planning to keep making it better by using even better data.

AI Revolution: Paralyzed Woman ‘Speaks’ via Digital Avatar

Researchers from UC San Francisco and UC Berkeley have developed a groundbreaking brain-computer interface (BCI) that empowers a woman with severe paralysis to communicate through a digital avatar. This achievement represents the first instance of generating speech and facial expressions directly from brain signals.

The system can convert these signals into text at an impressive speed of almost 80 words per minute, outperforming existing technologies. The BCI deciphers brain signals into synthesized speech and facial movements, enabling paralyzed individuals to communicate more naturally. Instead of recognizing whole words, it identifies phonemes, the building blocks of speech, which enhances both speed and accuracy.

The avatar's voice is tailored to mimic the user's voice before her injury, and facial animations are driven by software interpreting brain signals for various expressions. The researchers aim to develop an FDA-approved system for speech communication via brain signals, marking a significant stride toward comprehensive communication restoration for paralyzed individuals.

Nvidia Revenue Doubles on Demand for A.I. Chips, and Could Go Higher

Nvidia, that big tech company from Silicon Valley, is crushing it right now with its chips for artificial intelligence (A.I.). You know, those things that make computers think smarter? Well, their sales are through the roof, having doubled in the recent quarter. They earned a whopping $13.5 billion! That’s way more than what they thought they'd get, and it's making Wall Street folks super excited.

But there's a catch: they can't make these chips fast enough. This could be a chance for other chip-making companies to swoop in. Companies like Intel and AMD, who've been having a bit of a rough patch lately, might see this as a way to get back in the game.

Everyone's talking about Nvidia’s new A.I. chip, the H100. But they’re so in demand that there might not be enough of them until 2024. Nvidia's boss, Mr. Huang, says they’re doing their best to make more of these chips ASAP.

Wand.app raises $4.2M to scale its AI-powered creative tool for artists

Wand.app just bagged $4.2 million to beef up its AI tool for artists, with O’Shaughnessy Ventures fronting the dough. Instead of AI going rogue and doing its own thing, Wand gives artists the steering wheel, letting them mold and shape their work.

Created by big brain Grant Davis in 2021, the app’s mission? Make sure artists’ visions aren't lost in translation. The Wand magic lets artists teach the AI their style, and the app turns rough sketches into polished ideas. It's got a test group now, but the plan is to open the doors to everyone soon. Wand stands out from the crowd, letting folks use it on phones, tablets, and soon, computers. And for the Apple junkies, yes, it's Pencil-friendly. The new money? That's for team upgrades, new tools, and to bring the magic to computers.

Lex raises $2.75M for its AI writing tool that helps writers get past blocks

Lex just scored $2.75 million in seed money, led by True Ventures, for its AI writing tool that helps writers beat the dreaded block. This ain't your grandpa's Word doc. Created by Nathan Baschez, who also kickstarted Every, Lex aims to add a dose of AI smarts to your writing process. This tool ain't just bells and whistles; it's got all the formatting gadgets you'd expect. But when you hit a wall, just hit CMD+Enter and boom! The AI spits out what might come next.

Writers can also pop questions into comments, like "Is this sentence too long?" or "Give me a snappy headline." So will this make us all lazy writers? Nah, sometimes the AI churns out junk, but it’s handy for breaking through writer’s block. Privacy? Baschez promises they ain’t snooping on your work for now and aims to keep it that way.

Film and TV studios post AI jobs with salaries up to $1 million as workers strike

Amidst strikes by writers and actors in the film and TV industry due to concerns over AI's impact on their work, major entertainment companies such as Disney, Netflix, Sony, and NBCUniversal are actively advertising for artificial intelligence (AI) experts. These companies have posted about 26 AI-related job openings recently, with salaries often exceeding $200,000 annually and even reaching up to $1 million.

Netflix alone has listed 18 AI-related positions, including a job offering $1 million to develop video games using AI. These positions vary in their connection to content creation, addressing backend AI integration, content generation, or support for content creators.

While creative professionals are striking over AI's threat to their livelihoods, companies are investing in AI specialists, reflecting their focus on long-term innovation. This move comes as generative AI gains momentum in content creation, while unions demand compensation and consent for AI-driven work. Amid the debate, the industry seems to be aligning its resources with AI capabilities, potentially reshaping the entertainment landscape.

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.