<aside> 🤍 This guide is meant to be a starting point for those who want to start using SD, findings from experiments and a possibility to learn more for everyone who is already experienced in SD. Inspired by Ethans incredible travellers guide to latent space for DD, with most of what you’ll read driven by the amazing community around this open source software.

</aside>

Resources

• pharmapsychotic: Overview of all tools related to latent diffusion
• SD Resource Goldmine: Everything SD you might ever need
• OpenArt Prompt Book: How to create a good prompt
• Lexica.Art: Search Engine for SD prompts & settings with results
• List of published SD models and embeddings
  • CivitAI: Subset of the above (but with pictures!)

Starting out: Stable Diffusion settings

A lot of this info is from Jon Stokes’ awesome quickstart guide to Stable Diffusion, give him a visit!

Seeds and Prompts

You’ve probably already messed around a bit with Stable Diffusion courtesy of the Stability AI team’s in-house cloud app, Dream Studio. If you haven’t, then stop what you’re doing and go sign up so you can follow along with the rest of this section.

Stable Diffusion takes two primary inputs and translates these into a fixed point in its model’s latent space:

1. A seed integer (the starting noise of the diffusion process)
2. A text prompt (the description of the image you want to generate)

The same seed and the same prompt given to the same version of Stable Diffusion will output the same image every time. In other words, the following relationship is fixed:

seed + prompt = image

(Interestingly, this fixed relationship between the seed, prompt, and output image means you can take a given Stable Diffusion output image and seed, and then run the model in reverse  to get the original text prompt. AFAIK, no release of this functionality has been made.)

One practical application for the seed + prompt = image equation is that by holding the seed constant you can then subtly tweak the prompt to iterate closer to the exact image you want.