DR.AI Google's new LLM No.1, Animate any photo in 1 click, Apply Pattern to object in MJ + 5 new AI tools
This week: Elon predicted AGI in 3 years, while Nvidia’s CEO predicted 5 years. Meta released a near real-time universal translator and SDXL Turbo was made to look slow with a new Stream Diffusion Turbo project creating: 150 image generations per second!
Design Industry AI News
Google Gemini the model they claim beats Open.ai
Google surprised us all by launching their Gemini early. One way it differs from Chat GPT which combines vision and audio models together, is Gemini has been designed to be multi-modal from the outset.?
Imagine an AI model that had all the senses from day one, that would be a big advantage to understanding the world. Gemini isn’t there yet but it does seem to understand visual input, video and images very well. Let’s be clear, in spite of Google’s hype and big claims this is about on par with GPT when it comes to text, but watch the video to see some impressive never-before-seen capabilities. It may seem like they finally caught up Open.ai but they’re one year late and Open.ai has not been sitting still for that time (Q*?), but the race is hotting up. Try Gemini Pro by visiting the Bard website although the more advanced vision capabilities won't be available until next year and only for enterprise customers initially.
EDIT:
Getty Images admits the tide is against them and partners with Runway ML
Getty started the year by suing Stable Diffusion. Just 12 months ago the legality of AI-generated was far from clear - frankly, it’s still not, but with more images now produced by AI than photographers in the entirety of the last 150 years, we can be sure it’s here to stay.
Getty were probably trying to hold on to their business model but by dragging their feet they now have to partner with Runway, a first mover, to have any chance of staying relevant. It is pretty indisputable the value of a massive high-quality library to use as training data - and this is something Pika Labs doesn’t have yet. I personally will be delighted when we can generate high-quality videos and not spend hours searching stock libraries. Although, let’s be honest at that point will we have a human in the loop?
Learn
How to animate still images using MagicAnimate
The world has seen enough TikTok dances but you’re much more creative than that. Scroll to the bottom of this thread and you’ll find a way to create your own original pose data and the Magic Animate tool that will allow you to take photos or original characters and make them move with AI
And if you’re lucky enough to have a Nvidai GPU you download MagicAnimate with one click using Pinokio to run programs locally
Claude 2.1 can read 150,000 words but is it better than chat GPT?
Knowing when to use one model over another and how to get the best out of Large Language Models will have you crushing the competition
Create intricately patterned objects using this Midjourney tip
Credit to Rory Flynn for sharing this sexy little Midjourney tip to transfer a pattern onto to object within an image. Tested and approved.
A.I. Tools
Leonardo Realtime Canvas
Leonardo is already a very capable txt-2-img model but now they added a realtime drawing canvas too. I love how this video hints at a new way to animate.
领英推荐
Meta’s Imagine
Meta has launched a dedicated website for its AI-powered image generator called Imagine. It’s based on the Emu model which had impressive editing capabilities, simply by writing what you want to edit. Only available in the US at the moment so can’t be sure if these abilities have made it to this model. I hope so.
Visual Electric
A generative AI-based image creation interface focused on the workflows of designers. The tool is free to use, with a cap of 40 image generations per day. A premium plan is available for $16 per month, offering better generation speeds, no limit on image creation, and a license to use the images commercially
Deepswapper
Don’t underestimate the power of a good Meme. DeepSwapper AI is a free and unlimited face swap service that offers high-quality face swaps in just a click.?
LMStudio AI
Run LLMs locally On Mac Silicon or Windows. Essentially this is the future, we’ll all have our own AI models running on local computers doing task for us. Don’t expect Chat GPT performance yet although smaller models like Orca 2 (13B) are becoming more efficient and competing with? models 5-6x larger.?
A.I. Video
Motion Scrapbook?
Experiment Mixing AI, 3D+2D and remixing a couple of brands. (contains strobing lights)
CREDITS: Anthony Gibsonn
Pika Labs 1.0 looks like a lot of fun
The new object inpainting and outpainting feature for video looks like a lot of fun, anyone else still on the waitlist gagging to get their grubby hands on this?
CREDITS: Martin Haerlin
Research Paper
Readout Guidance: Learning Control from Diffusion Features
Readout Guidance, a method for controlling text-to-image diffusion models with ‘learned signals’. Basically in a variety of different ways you you can gain fine control over diffusion models to recreate poses using stick figures or drag and manipulate the image. Hard to explain just check out this thread for the videos:
Thanks for reading please share this with a friend who would appreciate it. If you prefer to read this in your inbox you can sign up here: Design Resources.AI
Peace