March 2024
May 2024
NVIDIA recently released FREE courses on AI.
Here are 6 best courses (with links) you don’t want to miss:
1. Generative AI Explained
URL: https://lnkd.in/d6vpQWE5
__________
2. Augment your LLM Using Retrieval Augmented Generation
URL: https://lnkd.in/dAhW_cER
__________
3. Building A Brain in 10 Minutes
URL: https://lnkd.in/d8g2wVSh
__________
4. Introduction to AI in the Data Center
URL: https://lnkd.in/dzMmRrJx
__________
5. Introduction to Networking
URL: https://lnkd.in/dXZiFDs2
__________
6. Accelerate Data Science Workflows with Zero Code Changes
URL: https://lnkd.in/dgxQrY_7
__________
If you like this, and want more AI courses, tutorials, and tools, join Superhuman, my AI newsletter with 600K+ subs now: https://lnkd.in/dEmezNQ2
A passion for creating spaces for AI Agents of change
As of the past two months, several significant updates and trends have emerged in the field of artificial intelligence (AI) globally:
16-Mar-2024: Midjourney Introduces Consistent Characters. Sort of…
By Christian Heidorn chris@tokenizedhq.com via m.convertkit.com
- now that Midjourney has finally released the one feature that so many people have been desperately waiting for.
- Consistent Characters!
- Similar to the style references that were rolled out a few weeks ago, this new feature is called “Character References”.
- It’s still early and there will undoubtedly be adjustments to how it works and what sort of images it produces.
- Nevertheless, I think it’s worth diving into this because as simple as it may seem on the surface, it’s more complex than I thought.
- My initial thoughts were “Wow!” but the more I use it, the more I am noticing its weaknesses too.
- So let me walk you through the basis of how it works and then I’ll share my thoughts on what I believe needs to improve.
-
- 1. Character References in a Nutshell
- Alright, so applying a character reference is pretty straightforward.
- First of all, you enter the prompt for the image that you’d like to create.
- Feel free to add all of the parameters that you’d typically want to use.
- Next, you need to add your character reference.
- Similar to how style references work, your character reference needs to be an image hosted on the internet.
- You can use an image that you’ve previously created in Midjourney but it doesn’t have to be.
- Here’s an example of how to use it:
- /imagine photo of a female police officer in new york city –ar 3:4 –cref <URL>
- Notice the new –cref parameter in there?
- That stands for “character reference” and it is followed by a URL to the image that you want to use.
- There is also a second parameter that comes with it: –cw
- –cw stands for “character weight” and if you don’t define it in your prompt, it will default to a value of 100 (the scale is 0-100).
- Remember, you don’t need to set this value unless you want to control the weight.
- Why would you want to do that?
- Well, because a value of 100 will not only use the facial features and hairstyle of your character, it will also transfer most of the clothing.
- Obviously, you don’t necessarily want that for every use case.
- Another side effect of a high –cw value is that the character features of reference will tend to dominate your prompt.
- For example, if your character reference has a very specific hairstyle or outfit, changing that via your text prompt becomes a lot more challenging at –cw 100.
- So let’s look at how different –cw values impact your images.
-
- 2. The Impact of Weights
- Understanding how the weights control the effect of your character reference is crucial.
- It can make or break your images, especially if you need to place your character in different locations and situations.
- Based on my initial experiments, there are 4 rough anchor points that you can use:
- –cref –cw 0: Only use facial features
- –cref –cw 25: Facial features and some hairstyle
- –cref –cw 50: Face, hair, and some parts of the outfit
- –cref –cw 100: Face, hair, and outfit
- These distinctions are important and you can see that in the images below.
- This illustrates quite well that characters, while extremely helpful, aren’t necessarily a one-shot solution.
- It will still require the right balance of reference images, weights, and also a strong text prompt to create precise images with your character.
- And speaking of challenges, it’s not all moonlight and roses.
- However, if I apply the style reference, things change quite dramatically:
-
- 3. Midjourney “Sees” Differently
- One thing that I have noticed throughout all my testing is that Midjourney seems to perceive my image references quite differently than I do.
- You may or may not have noticed this in my examples above.
- Although it uses my reference image to create a consistent character throughout all of its generations, it still doesn’t quite look like my original reference.
- And that’s even though the image originated from Midjourney in the first place.
- I’ve noticed that it tends to exaggerate certain facial features, such as the mouth, jaw, nose, eyes, and even freckles.
- In the specific example above, my reference did not have such a wide jaw, nor were her lips as puffy and the freckles are just over the top.
- In other tests, Midjourney would give my character an extremely pointy nose, wide puffy lips, and yet again, a wide jaw and way too many freckles.
- So that’s why I say that Midjourney “sees” differently, or at least differently from how I see things.
- Once it’s decided on how to interpret your reference, it will generate strikingly consistent images.
- However, they’re not necessarily a picture-perfect representation of your reference.
- And honestly, that’s a little disappointing.
- But hey, this feature has only been around for a few days now and I’m sure it will go through multiple iterations of improvements.
- Until then, this is already a huge step forward and I can’t wait to see what people create with these features.
- Applications include climate modeling, carbon footprint analysis, energy optimization, and the development of AI-powered tools for renewable energy production and consumption optimization.
Advancements in Natural Language Processing (NLP):
- Continued progress in NLP models, with innovations such as more efficient architectures, enhanced contextual understanding, and improved multilingual capabilities.
- Transformer-based models like GPT (Generative Pre-trained Transformer) series, BERT (Bidirectional Encoder Representations from Transformers), and their derivatives remain at the forefront, achieving state-of-the-art performance in various language understanding tasks.
Ethical AI and Responsible AI Adoption:
- Heightened focus on ethical considerations and responsible AI deployment, with organizations emphasizing transparency, fairness, and accountability in AI systems.
- Increased scrutiny and regulation surrounding AI technologies, particularly concerning bias mitigation, data privacy, and algorithmic transparency.
AI in Healthcare:
- Continued integration of AI-driven solutions in healthcare, including diagnostic assistance, patient monitoring, drug discovery, and personalized treatment recommendation systems.
- Advancements in medical imaging analysis, natural language processing for clinical documentation, and predictive analytics for disease management.
AI and Climate Change Mitigation:
- Growing emphasis on leveraging AI for climate change mitigation and environmental sustainability efforts.
- Applications include climate modeling, carbon footprint analysis, energy optimization, and the development of AI-powered tools for renewable energy production and consumption optimization.
AI in Finance and Fintech:
- Increased adoption of AI technologies in the finance sector, particularly in areas such as risk assessment, fraud detection, algorithmic trading, and customer service automation.
- Development of AI-powered chatbots and virtual assistants to enhance customer experience and streamline financial transactions.
Autonomous Vehicles and Robotics:
- Continued progress in autonomous vehicle technologies, with advancements in perception systems, decision-making algorithms, and real-time mapping capabilities.
- Expansion of AI-driven robotics applications in industries such as manufacturing, logistics, agriculture, and healthcare, with a focus on improving efficiency, productivity, and workplace safety.
These trends reflect the continued evolution and diversification of AI applications across various sectors, driven by ongoing technological advancements, increased awareness of ethical considerations, and growing demand for AI-driven solutions to address complex societal challenges.
Where do we go from here? How fast? What are the implications?
Ethics and architecture: podcast (coming soon!)
- Collaborate with fellow architects.
- Showcase your projects.
- Experience the world coming to life://AI.
Bot2Billions Newsletter (subscribe here)
- A world of thought-provoking articles.
- Case studies that celebrate architecture.
- Exclusive access to design insights.
Hey there, you Fearless one (click for a jolt).
Michael Fenech
Join ∞+ subscribers
Stay in the loop with everything you need to know.