Tons to talk about in the AI news world. Here is a short, condensed article of the highlights. Follow links for testing and more information, and no, we do not have any affiliation with these companies.
Daniel MacDougall
If you enjoy content like this, signup to our free newsletter for AI news and tools.
Luma AI Dream Machine Release
The Luma AI Dream Machine has been released and is available for personal use. It competes well with other video creators, particularly excelling in image-to-video tasks, though it still struggles with text-to-video prompts. Users had previously complained about long wait times, but these have significantly improved.
Text-to-video functionality still requires refinement, but users can experiment with up to 30 free videos per month. For more detailed information check this out, or visit Luma Labs Dream Machine.
Stable Diffusion 3 Medium
Stable Diffusion 3 Medium is now publicly accessible. This text-to-image AI, using a 2 billion parameter SD3 model, claims to deliver better color, lighting, and photorealistic outputs compared to previous versions.
However, it requires more detailed prompting compared to other models. If you're just starting out with text-to-image AI, I would suggest something like ChatGPT or Leanoard AI for a more conversational approach to image creation, but if you've previously tried Stable Diffusion and it didn't quite cut the mustard, now might be the time to give it a second chance.
For more details, visit Stable Diffusion 3 Medium on Hugging Face.
Leonardo AI's Phoenix Model
Leonardo AI has announced its new Phoenix model, which offers superior image quality, better prompt adherence, and coherent text in images.
This improvement is notable as Midjourney still struggles with text coherence. After testing the Phoenix model, its performance is impressive enough to pull me away from Midjourney, and I use Midjourney quite a bit. For more information, visit Leonardo AI.
They also have a great free plan, but like most free plans, there are limitations.
Midjourney Personalization Option
Midjourney has introduced a personalization feature, allowing users to train the AI on their unique style by selecting AI-generated images. While the selection process can be exhausting, the resulting style code provides personalized outputs with a distinctive personal edge. This allows more consistency between your image generations. Something the community has been asking for quite some time.
To set this up go to midjourney.com, select tasks, and then rank images. The more images you rank the better the AI will be at depicting your style.
Apple's WWDC 2024 Keynote
At their annual WWDC 2024 keynote, Apple announced they are integrating AI, termed "Apple Intelligence," into pretty much anything. This includes - a very short summary:
AI image generation on devices
AI assistance built into the phone
AI upgrades to Siri
AI writing assistance
There is too much to discuss here. Read this article for the details.
OpenAI and Apple Partnership
OpenAI and Apple have formed a partnership without financial transactions, focusing instead on mutual exposure.
Also, ChatGPT will not run on Apple's OS, and OpenAI won't be able to detect IP addresses while using ChatGPT on Apple devices, so they say. But this merger is quite interesting given the partnership between OpenAI and Microsoft and now entry into Apple.
DeepMind and Harvard's Virtual Rat Brain
DeepMind and Harvard have created an artificial AI brain for a virtual rat, capable of realistically controlling a rat in a simulation. This involved creating a biomechanical model of a rat's body in a simulator named MuJoCo, ensuring the virtual rat obeyed physical laws and moved naturally.
DeepMind's AI then used inverse dynamics modeling to train a neural network with data on real rat movements. This neural network could generalize behaviors and movements, closely matching the neural activity patterns of real rats.
This advancement is a significant step towards developing AI robots and animals that mimic real-life behaviors and movements, raising the prospect of a future where AI robots and real people become indistinguishable.
A first step forward in second guessing if a robot is walking down the street or a real person. If we get to this point, I think all realistic bots should have a name tag.