Luma AI has recently released their new video generation tool, Dream Machine, which aims to compete with other AI-powered video creation platforms like Soar, Veo, Cling, Pika, and Runway. While the tool shows promise in certain scenarios, it still has some limitations that users should be aware of.

One of the main issues with Dream Machine is the long wait times for video generation, especially during periods of high demand. In the early days, some requests took up to 7 hours to start processing, which can be quite frustrating for users. Luma has since scaled up their infrastructure, but the wait times can still be significant.

In terms of the quality of the generated videos, Dream Machine struggles with text-to-video generation. The examples provided in the transcript show that the tool has difficulty accurately depicting elements like a wolf howling at the moon or a monkey on roller skates. The generated videos often have inconsistencies, such as missing limbs or incorrect positioning of objects.

However, where Dream Machine seems to shine is in the image-to-video feature. The transcript showcases several examples of this, including a colorful futuristic city, a pixelated video game wolf house, and a cabin in the woods. These image-to-video conversions appear to be more realistic and coherent than the text-to-video attempts.

It's worth noting that Dream Machine is currently in a research preview stage, and users can generate up to 30 videos per month for free. After that, the pricing model is around $0.25 per video. As the tool continues to evolve, it will be interesting to see if Luma can address the current limitations and improve the overall quality and consistency of the generated videos.

Overall, Dream Machine shows promise, but users should approach it with realistic expectations, especially when it comes to text-to-video generation. The image-to-video feature appears to be the stronger aspect of the tool at the moment.

Stable Diffusion 3, the latest iteration of the popular open-source AI image generation model, has finally been made available to the public. Let's take a closer look at what this new version has to offer.

Improved Text-to-Image Capabilities

One of the key improvements in Stable Diffusion 3 is its enhanced ability to incorporate text into the generated images. The model now seems better at translating textual prompts into coherent and detailed visual representations. This can be seen in the examples provided, where the text-based prompts result in more accurate and visually appealing images.

Prompt Engineering Still Required

However, it's worth noting that Stable Diffusion 3 still requires a certain level of prompt engineering to achieve the best results. While the model has improved, users may need to provide more detailed and specific prompts to get the desired outcomes, especially for complex or detailed images. This is in contrast to some other AI image generation models that can produce high-quality results with more straightforward prompts.

Inconsistent Quality

The quality of the generated images can also be somewhat inconsistent. While the model is capable of producing impressive results in certain scenarios, such as the "astronaut in a jungle" example, it still struggles with simpler prompts like "a monkey on roller skates." This suggests that Stable Diffusion 3 may not yet be at the level of some of its competitors in terms of overall image quality and consistency.

Continued Advancements Needed

Overall, Stable Diffusion 3 represents a step forward in AI image generation, but there is still room for improvement. As the technology continues to evolve, we can expect to see further advancements in the model's ability to translate text into high-quality, coherent images without the need for extensive prompt engineering. The community's ongoing efforts to refine and enhance Stable Diffusion will be crucial in driving these improvements.

Full disclosure, I am an adviser for Leonardo, but they have zero control over what I say. If something is funky about it, I'm going to point it out. Being an adviser to them does not impact what I actually say about them.

That said, Leonardo just released a new custom model called Leonardo Phoenix. This is their own foundational model, not a version of stable diffusion. The major features of this new model are:

• Enhanced prompt adherence - It can better understand and adhere to the prompts you provide.
• Coherent text in images - It can incorporate text into the images in a more natural and coherent way.
• Superior image quality - The generated images are of higher quality compared to previous models.
• More creative control - You have more control over the creative direction of the images.

However, some features like image guidance, elements, and photorealistic versions aren't available yet. They're still working on implementing those extra features.

Let's take a closer look at the model in action. I'll go to the Leonardo website, select the Leonardo Phoenix preset, and try a simple prompt - "a wolf howling at the moon".

Here are the images it generated:

[Image 1] [Image 2] [Image 3] [Image 4]

I don't know about you, but these are quite a bit more impressive than what I just saw out of stable diffusion 3. The model seems to have done a great job of understanding the prompt and creating coherent, high-quality images.

Let's try another example - "a penguin holding up a sign that says Mr eow".

[Image 1] [Image 2] [Image 3] [Image 4]

The text is spelled correctly in every image, and the penguin holding the sign looks pretty good. The model handled the text integration very well.

Overall, the Leonardo Phoenix model feels like a step up from stable diffusion 3. I highly recommend playing with both and seeing which one works best for your needs. The enhanced prompt adherence and text integration capabilities of the Leonardo Phoenix model are particularly impressive.

Sonno, the AI-powered music creation platform, has recently unveiled a groundbreaking feature that allows users to transform their own audio recordings into fully-fledged songs. This innovative capability, available to Sonno Premium subscribers, empowers creators to harness the power of AI to elevate their musical ideas.

Here's how it works:

1. Record or Upload Audio: Users can either record audio directly within the Sonno platform or upload an existing audio file. This could be a simple guitar riff, a vocal melody, or any other musical snippet.

2. Extend and Enhance: Once the audio is uploaded, users can select the "Extend" option. Sonno's AI-driven algorithms then analyze the input and automatically generate an extended, fully-produced song, complete with additional instrumentation, harmonies, and lyrics.

3. Customization Options: Users have the ability to further refine the generated song by adjusting parameters such as the genre (e.g., acoustic pop, electronic, etc.), the inclusion of a beat, and the generation of random lyrics.

The results are often surprisingly impressive, with Sonno's AI seamlessly blending the user's original audio with its own musical compositions. The generated songs maintain the essence of the user's input while elevating it to a professional-sounding level.

This feature opens up new creative possibilities for musicians, songwriters, and hobbyists alike. Users can experiment with different ideas, quickly turn sketches into complete compositions, and even collaborate with the AI to bring their musical visions to life.

As Sonno continues to refine and expand its capabilities, this audio extension feature is poised to become an indispensable tool in the arsenal of modern music creators. By empowering users to transform their raw ideas into fully-realized songs, Sonno is redefining the way we approach music production and composition.

Apple made a huge push into AI at their recent WWDC event, integrating AI capabilities across their entire ecosystem of devices and services. Here are the key highlights:

AI in iOS, iPadOS, and macOS

• Apple is building its own AI and integrating it deeply into iOS, iPadOS, and macOS. This includes features like:
  • Proofing, rewriting, and summarizing text in apps like Notes, Mail, and more
  • AI-powered vision capabilities in apps like Notes and Calculator to analyze images and handwriting
  • Prioritizing and summarizing emails and notifications using AI

Image Playground

• Apple's new image generation feature, called "Image Playground", allows creating illustrations, animations, and sketches using AI.
• It has a unique interface where you can see the different contexts the AI will blend together.
• The AI is limited to non-realistic styles to avoid deepfakes.

GenEmoji

• Users can create their own custom Emoji using AI, which can then be used as reactions and stickers.

Siri Improvements

• Siri can now be used by typing instead of just voice.
• Siri will leverage on-device and iCloud-based AI, only using OpenAI's ChatGPT when it can provide better answers.

Apple Cloud and OpenAI Partnership

• Apple is building a secure "Apple Cloud" to handle sensitive AI processing.
• They are also partnering with OpenAI to allow Siri to leverage ChatGPT when appropriate, with user permission.

Overall, Apple is deeply integrating AI across its entire product lineup, leveraging its own technology as well as strategic partnerships. This represents a major push to make AI a core part of the Apple experience.