STABLE DIFFUSION OVERVIEW
=========================

What's Stable Diffusion?
------------------------

Open-source AI image generation from Stability AI. Unlike DALL-E or
Midjourney, they released the weights and code publicly back in August
2022. Anyone can run it on their own hardware.

That openness changed everything. People started downloading it, studying
how it works, fine-tuning it, and sharing what they made. A whole
ecosystem popped up almost overnight.


Stable Diffusion vs AUTOMATIC1111
---------------------------------

These are NOT the same thing:

  Stable Diffusion = the AI model itself (the brain)
  AUTOMATIC1111 (A1111) = a web interface to control it (the dashboard)

A1111 is a community project, NOT made by Stability AI. It's just the
most popular way to use Stable Diffusion. Other interfaces exist:

  - ComfyUI (node-based, more technical)
  - Forge (A1111 fork, optimized for speed)
  - InvokeAI (polished UI, different workflow)

Tonight we're using A1111 because:
  - Most documentation assumes A1111
  - Most model pages give A1111 settings
  - Biggest community, most extensions
  - Beginner-friendly interface


The Open Secret: Training Data Reality
---------------------------------------

Here's something you should know upfront. Cloud AI services like DALL-E
and Midjourney heavily filter their outputs. You type a prompt, you get
a clothed person. That's by design.

Local Stable Diffusion models? Different story. Many community models
are trained on uncensored datasets. Stability AI maintains plausible
deniability, but the reality is: these models learned from everything.

What this means practically:

  - Default outputs may not include clothing unless you ask for it
  - You'll likely need negative prompts like "nude, nsfw" to get
    safe-for-work results consistently
  - This isn't a bug, it's how the training data shaped the models

None of this is a value judgment. Just know what you're working with.
If you're demoing at work or showing family, test your prompts first
and use appropriate negative prompts. The tutorial covers this in the
negative prompt section.


Model Generations
-----------------

SD 1.x (2022)
  Original release. Makes 512x512 images. Runs on consumer GPUs.
  Good for 2022, showing its age now.

SD 2.x (Late 2022)
  Better quality but broke compatibility with community tools.
  Nobody really adopted it.

SDXL (2023) - THIS IS WHAT WE USE
  Current generation. Native 1024x1024. Way better at composition,
  anatomy, and text. Models are bigger (~6.5GB vs ~2GB) but the
  results speak for themselves.


Why Open Source Matters Here
----------------------------

Because Stability released everything openly, the community built
specialized versions. Want anime? Someone trained a model for that.
Photorealistic portraits? Got it. Western comics? Multiple options.

The models in this tutorial are community fine-tunes built on SDXL,
each dialed in for a different look.


Links
-----

A1111 repo: https://github.com/AUTOMATIC1111/stable-diffusion-webui

WebUI runs at: http://127.0.0.1:7860