Stable Diffusion (2024): Essential Parameters You Need to Know (CFG and More)
- Jakob Stewart
- Aug 2, 2024
- 1 min read
Hi, I'm Jakob Stewart, a digital artist working with NSFW AI in adult content creation.
In this article, I will talk about the main parameters that you need to know to use Stable Diffusion.
Click the links below to go directly to the section:
What is CFG in Stable Diffusion?
The Stable Diffusion CFG defines how much the model considers the prompt to generate the final image.
The most common mistake beginners make on Reddit is setting the CFG (Classifier-Free Guidance) too low or too high.
Too high: The image becomes overly saturated.
Too low: The image doesn't follow your prompt closely enough.
Ideal range: A CFG between 3 and 8 is usually best.
⚠️ Some hyper, turbo, and lightning checkpoints might work with lower CFG.
Always read the model descriptions on Civitai to find the right parameters for each checkpoint.
What are the Sampling Steps in Stable Diffusion?
Sampling Steps are the number of steps that will be executed to generate your image.
For most use cases, the ideal number of steps is between 30 and 60.
This helps in achieving a good balance between detail and speed.
⚠️ Some hyper, turbo, and lightning checkpoints might work with lower CFG.
Always read the model descriptions on Civitai to find the right parameters for each checkpoint.
RealisticVision and RealVis, work very well with less CFG and fewer steps.
What About Checkpoints? The Best Stable Diffusion Model
I often see people using the default Stable Diffusion model:
v1-5-pruned-emaonly.ckptTo improve your results, use an appropriate checkpoint.
I will write a guide on checkpoints and LoRAs soon.
Subscribe for FREE to be updated.
