Welcome to Issue 2 of unfolding ai, we are delighted to have you as a reader

In this issue we are going deeper into the world of generative AI, with a focus on getting started in the visual. Looking into what are the current tools we are using, some alternatives and how you can get hands on.

One of our key beliefs at unfold:ai is that creativity and making is the unlock to a greater life wellbeing, fun and enjoyment. But like every skill it needs to be exercised, the good news with AI by your side everyone can unleash the creative ideas inside themselves.

Just a warning, these products come with subscriptions beyond the ‘free’ and starter plan, there is overlap so perhaps master a few tools, rather than them all.

We hope you find this newsletter informative and valuable, and we look forward to hearing your feedback.

It’s in depth this week, make sure if you are a subscriber you read it all!

Best regards,

Paul, Co-founder (and newsletter editor)

p.s As this is a brand new newsletter, we would really appreciate it if you spread the word and did a social share. The icons are at the top of the newsletter,

What is Generative image

Generative Image is a AI model which has been trained on millions of images which have meta data (descriptions). This is combined with a text processing AI Model which converts entered text to image. As a solution it has been around a long time, Google Dream was first released in 2014.

Leaping forwards to 2022, systems like clip-diffusion, disco-diffusion paved the way for the text to AI solutions we have today. The predominant headline tool since its release has been Midjourney. Although it has a slightly complex registration process using discord and its own services, its still considered the go-to tool by many.

It is a little like using your camera on a pre-set ‘scene’ mode, you have plenty of control. But behind the generate Midjourney is providing a large amount of reinforcement trained processing, which generates it’s ‘look’.

The second solution is stable diffusion. Stable diffusion is often the underlying technology for 3rd party sites. Some of these are targeting the mobile app content creator, some have much more (insanely more) complex interfaces to allow the digital artist to get deep underneath, camera raw mode (to continue the camera analogy). Until very recently (this week for version 1.0) it was behind Midjourney on ease of output. That’s all changed the new SDXL Model is a step change.

Many of the features you see in Midjourney, or TikTok apps, are driven by developments which started in stable diffusion.

There really is no excuse not to have a go with these tools. Here is an example prompt to get you started, will work in any text to image generator. If you are stuck for inspiration, all the tools have discord groups, or Facebook pages, or Instagram.

There is an text to image AI feature in just about everything now including, LinkedIn, Canva, Photoshop & Power Point.

Here is a prompt to try anywhere.

a beautiful landscape picture, distant trees, mountains, clear azure sky, tranquil, high detail, golden hour light

Here are a few links for you to explore

Genesis, the trailer

Last week we talked a little about Nicholas Neubert’s trailer Genesis. He kindly shared his approach with everyone.

It is such an interesting approach that there is a more detailed breakdown for premium subscribers but here is an asset breakdown. Construction time was 7 hours, remember though he is already a skilled user of all these tools.

• 316 Prompts in Midjourney

• 128 Upscales in Midjourney

• 310 Runway ML Videos

• 1 Video (generated text to video)

• 44 Runway Videos in the clip

Regulations and AI, quick update

The Biden Administration (America) has received a commitment from 7 of the leading AI firms to some voluntary constraints. Recognising that regulation is going to run behind pace of change. This combines with the ‘Frontier Model Forum’. Between them these are the key points.

What they have agreed to do in combination:

• Adhere to principles of safety, security and trust in AI systems

• Test AI systems thoroughly before release, protect model weights

• Label AI-generated content, report on AI capabilities and limitations

• Prioritize research on mitigating AI risks like bias and deception

• Develop AI to address major societal challenges

• The Frontier Model Forum will convene companies, governments and civil society - It will develop best practices, advance AI safety research, and enable information sharing

• The Biden administration will develop an Executive Order and pursue legislation to regulate AI - They will collaborate with international allies to align AI governance standards

Who is signed up

• The Frontier Model Forum includes Anthropic, Google, Microsoft, OpenAI

• The Biden administration commitments involve Amazon, Anthropic, Google, Inflection, Meta, Microsoft, OpenAI

As of writing conspicuous by its absence is X.ai (Elon Musk)

Quick things to check out

40 years of robots at Boston Dynamics

Did you see this video out of context ?

I saw this video of just the robot playing table tennis, at very first glance it looks convincing. Then as you watch hopefully, your critical mind began to as the question is this real?

A prompt to try in chatGPT (and other LLMs)

I want you to solve this problem, show your working and explain each step of your approach, and why.

You need to explain how to stack the following items on top of each other without breaking any of them. It must be stable when finished.

A laptop
A book
9 Eggs
A bottle of water
A nail

Results do vary. The actual tip here for reuse though is just this bit.

I want you to solve this problem, show your working and explain each step of your approach, and why.

Did you miss these articles on our blog?

Our Events

Go Premium on us

Curious what a premium subscriber gets to discover. Try it for a month on us.

Subscribe to Premium to read the rest.