Vahin Parmananda

[Vahin Parmananda @Vahin]

This is the best tweet I’ve ever seen.

[“paula” @paularambles]

italian grandmas? you mean the penne gesserit

[Vahin Parmananda @Vahin]

@zachtratar I like the sound of this but…

[Vahin Parmananda @Vahin]

“The downfall of the great is the joy of the mediocre”
I don’t actually like Lisandro but that quote goes way too hard - I love it.

[All About Argentina 🛎🇦🇷 @AlbicelesteTalk]

🚨❤️ Lisandro Martínez: "We faced adversity, achieved what once seemed impossible, and never looked for excuses, we only found reasons to keep going.

"We defended our colors far beyond the pitch, and shared unforgettable moments with the national team, alongside our families and with the unconditional support of every Argentine. It will always be an immense honor to represent you.

"Remember: “The downfall of the great is the joy of the mediocre.”

Thank you all so much. Take care of our country.

I love you Argentina."

[Vahin Parmananda @Vahin]

@WaynaWaver @dielawn714 @ClaudeDevs Exactly. Seems like the solution is right there.

[Vahin Parmananda @Vahin]

@qmr55 @ClaudeDevs Such a bummer. Guess I need to get a MacBook after all.

[Vahin Parmananda @Vahin]

@NotionHQ is insane. I’ve been spending a week trying to figure out how to build a second brain in the system @karpathy described while still using Notion, and they go and make my life so much easier! 🙏

[Notion @NotionHQ]

.md, meet Notion 👋

You can now open Markdown files in Notion as a read-only preview, then import them as Notion pages.

[Vahin Parmananda @Vahin]

@mikefutia TRENDS

[Vahin Parmananda @Vahin]

@buildanythingso 👋

[Vahin Parmananda @Vahin]

@NotionHQ Notion AI computer use?

[Vahin Parmananda @Vahin]

@benln Yes please 🙏

[Vahin Parmananda @Vahin]

@araseb_ Conspiracy

[Vahin Parmananda @Vahin]

Diffusion Interactive Cheat Sheet

https://t.co/GcR8VwzfuR

[Vahin Parmananda @Vahin]

One of the most common questions in AI:

“How does generative AI work?”
When you are immersed in it, you develop an intuition. Explaining it clearly—without losing people halfway through—is much harder.

In architecture and design, image generation tools such as Midjourney, Stable Diffusion, Nano Banana and Adobe Firefly are now part of everyday workflows. With AI built into tools like Excel, WhatsApp and Outlook, this is no longer niche—it is mainstream.

So let’s break it down.

How does it work?

Many of these tools rely on a type of AI called latent diffusion models (LDMs).
This may sound technical, but it does not need to be.

Let’s break it down step by step.

---

L for latent

The key difference between these models and earlier approaches is that they do not work directly on full images.
Instead, they operate in what is called latent space.
Think of it like this:

Imagine you have hundreds of photos of your pet.
Each image is highly detailed—every pixel adds complexity. In technical terms, this is a high-dimensional space.

Now imagine compressing these images into a simplified “map” that still captures the important features:

shape, colour, texture, patterns

This simplified map is latent space.
Instead of working with millions of pixels, the AI works with a compressed representation of meaning.
Your words are handled separately.
The prompt “small fluffy brown dog with big eyes” gets translated into something the model can match against—the idea, not the pixels—and that steers what it builds in latent space.

---

D for diffusion
This is where the process becomes more intuitive.

Diffusion is the process of adding and removing noise.

A useful analogy:
If you remember tuning an old TV signal, you would see static first—then the image would gradually come into focus.

Diffusion models work in reverse.

Step 1: Start with noise
The model begins with what looks like pure randomness—visual static.

Step 2: Gradual refinement
Step by step, the AI removes noise and reshapes the image based on your prompt.

Step 3: Image emerges
After many iterations, the noise resolves into a coherent image.

Important clarification

The model is not “finding” an image hidden in the noise.
It constructs the image step by step using what it learned during training.

---

M for models
A model is simply:

A system trained to recognise patterns and generate outputs.
For image generation, models train on:
large datasets of images

paired text descriptions
Over time, the AI learns relationships such as:
what a “puppy” looks like

what “wearing a hat” means

how styles and compositions work

So when you type:

“a puppy wearing a hat”
…it generates a new image based on learned patterns. It generally does not copy—it recombines.

---

What's that noise?

Earlier, we noted that the process starts with noise.
This noise really is random—sampled fresh each time.

It is simply the blank canvas the model starts from, and all the structure appears later, as the model reshapes that randomness step by step.

From there, the model performs reverse diffusion:
interprets the prompt
identifies patterns
progressively reshapes the noise

Until:
shapes emerge
colours align
textures form
—and the final image appears.

---

Why do results change each time?

Even with the same prompt, results usually differ.
Each generation starts from a different seed.
A seed is a number that determines the initial noise pattern.

same prompt + same seed → identical result

same prompt + different seed → different result

Think of it as starting from a different sketch each time.

---

Final thought
That’s it — that’s the noise with LDMs, and how your favourite generative AI tools create magic from chaos. The next time someone asks you about AI image generation, there will be no confusion about diffusion!
The next time someone asks how AI image generation works, you can explain it clearly—without the jargon.
But wait, there’s more!

I’ve created a one-page interactive cheat sheet explaining how diffusion works in the comments below.

[Vahin Parmananda @Vahin]

@yuzu_jpg @X Hi Yuzu 👊

[Vahin Parmananda @Vahin]

HTML Blocks are insane!
Here I gave it a simple prompt to build an interactive globe showing AI legislation around the world.

[Notion @NotionHQ]

New block in Notion: HTML.

Build interactive HTML right on your Notion page. Ask AI to turn your content into interactive explainers, prototypes, or diagrams.

Share with your team to use and tinker together.

[Vahin Parmananda @Vahin]

This is sick!

[NASA's Kennedy Space Center @NASAKennedy]

The planet can spell your name – literally. 🔤🌍

This Earth Day, see your name written in landscapes captured by Landsat: https://t.co/kcP12dhsI2

[Vahin Parmananda @Vahin]

@onuro Keen to see the prompt

[Vahin Parmananda @Vahin]

@NotionCoach @NotionHQ Agent

[Vahin Parmananda @Vahin]

@nrqa__ X