QUICK REFERENCE - BOTH PLATFORMS
=================================

Platform Comparison
-------------------

+---------------------+------------------------+------------------------+
| Item                | macOS Apple Silicon    | Linux (Ubuntu)         |
+---------------------+------------------------+------------------------+
| Python version      | 3.10 (via Homebrew)    | 3.10 (native or PPA)   |
| Package manager     | Homebrew               | apt                    |
| Key dependencies    | cmake, protobuf, rust  | libgl1, libglib2.0-0   |
| Launch script       | ./webui.sh             | ./webui.sh             |
| Essential flags     | --skip-torch-cuda-test | --xformers (NVIDIA)    |
|                     | --no-half-vae          |                        |
| Access URL          | http://127.0.0.1:7860  | http://127.0.0.1:7860  |
| First run time      | 10-30 minutes          | 10-40 minutes          |
+---------------------+------------------------+------------------------+


Useful Flags
------------

+----------------+--------------------------------------------------+
| Flag           | What it does                                     |
+----------------+--------------------------------------------------+
| --autolaunch   | Opens browser when ready                         |
| --medvram      | Cuts VRAM usage (use if you're running tight)    |
| --lowvram      | Even more aggressive - slower but works          |
| --listen       | LAN access (binds to 0.0.0.0)                    |
| --port XXXX    | Use a different port than 7860                   |
| --api          | REST API for scripting                           |
| --xformers     | Memory-efficient attention (NVIDIA only)         |
| --no-half      | Full precision - slower but more compatible      |
| --no-half-vae  | Fixes black/corrupted images                     |
+----------------+--------------------------------------------------+


First Run Notes
---------------

First launch pulls down about 9GB of stuff. Go grab a coffee - it'll take
10-40 minutes depending on your connection.

Models go in: models/Stable-diffusion/

You don't need the brkirch Mac fork anymore. Those optimizations got merged
into the main repo.


PNG Metadata: Images Are Their Own Database
-------------------------------------------

Every image you generate contains all the metadata needed to recreate it.
The PNG file stores:

  - Full prompt and negative prompt
  - Seed number
  - Sampler and steps
  - CFG scale
  - Model name
  - Image dimensions

To reuse settings from any image:

  1. Go to the PNG Info tab in the WebUI
  2. Drag and drop your image (or click to upload)
  3. Click "Send to txt2img" (or img2img)
  4. All parameters auto-populate

This means you can share images and others can reproduce your exact results.
It also means your workflow is self-documenting - no need to keep notes
about what settings worked.

Note: Some image editors strip metadata when saving. If you edit in Photoshop
or similar, export as PNG and check "preserve metadata" or similar option.