Futuro Fortissimo Archive
🎼 ff.40 5 applications of AI
A Pokémon for every nation and cakes in every flavour +1
Published · Michele Merelli
- Can it create a Pokémon for every nation in the world?
- Will it automate making PowerPoint presentations?
- AI that speeds up the development of AI?
Themes🤖 AI, agents and models
🦄 ff.40.1 Gotta catch ‘em all
Substack ↗Stable Diffusion is an AI image generation technique that's showing interesting results. The output (an image generated from text) is similar to DALL-E, but the approach is different (a somewhat technical blog on it).
Starting from a generic model, you can “train” the algorithm on a subcategory of infinite styles. For example, Pokémon. So we can ask what Obama, Trump or Girl with a Pearl Earring would look like in the Kanto region.
Someone wondered what the mascot Pokémon would be for every nation in the world. Some examples:
Ready for Italy? Here it is:
Someone should do an artistic analysis to extract the symbolism the AI condensed into the creature. The green of the flag, the gold of Roman opulence, the symmetry of art. Maybe.
🔮 ff.40.2 A parallel future of design
Substack ↗Back to the future. On Instagram, An Improbable Future uses similar techniques to generate objects that mix the analogue world, futuro fortissimo and improbable design. I find them evocative: they open onto a parallel world, almost a Platonic hyperuranion.
Let's remember once again that these images are generated simply from a text description of what the artist wants to see, including the render-like style.
👨🏫 ff.40.3 PowerPoint presentations
Substack ↗Want (again starting from text alone) a PowerPoint presentation with themed AI-generated images and an AI-animated narrator? Here's the guide on Twitter:
fabians. eth@fabianstelzerFully AI-generated powerpoint presentations are even more fun when you add an AI narrator (using@synthesiaIO) here
Again, the GPT-3 prompt generates *everything* just based on inputting:
"How To Survive In The Wilderness: A Practical Guide" in the style of 80s illustrations1:42 PM ∙ Aug 24, 202244Likes4Retweets
🍰 ff.40.4 Cakes in a thousand flavours + 1
Substack ↗Fraternal twins. As mentioned, DALL-E and Stable Diffusion results are similar; let's look at an example of the subtle differences.
We can ask the AI to change the style or an element of the image. DALL-E did this by asking the user to “paint” over the areas to change.
A Google team proposed a less invasive editing method (it doesn't require inpainting but starts from editing just the text describing the scene, Prompt-to-Prompt Image Editing):
In the previous study, fidelity to the starting image was a problem (here trying to get different cakes, with different ingredients or materials):
Every flavour + 1. Thanks to the latest implementations (and by working on the “attention” the model pays to certain details) the result is better and fortissimo:
📈 ff.40.5 Predicting AI with AI
Substack ↗Artificial intelligence is also being used to predict the future of artificial intelligence. It's getting Hegelian, don't you think? The result is, needless to say, exponential.
👋 And that's all for this fortissimo week! 🔜 In the next ff we'll talk with GPT-3! See you soon, Michele
Want to support this project? Subscribe to the newsletter and share it with a friend on WhatsApp
👇🔥Subscribe
Or buy me a coffee 😏
☕
And remember to leave a little heart below.
❤️