How many refinement passes the model makes while turning noise into the image.
What the AI does
More steps usually add detail up to a point, but take longer; beyond that, changes are small.
When to use it
Lower for quick drafts, moderate for finished images.
Where it works
In most Stable Diffusion apps (for example AUTOMATIC1111 or ComfyUI) this is a setting next to the prompt, not something typed in it. ChatGPT, Gemini and AI Compass have no such control.
Common mistakes
Setting very high steps and expecting much better images.
Comparing step counts across different samplers.
Example
Explain how the number of sampling steps affects speed and detail in Stable Diffusion, and suggest a sensible range for drafts and final images.
CFG scale
Also written: guidance scale, CFG, guidance
How strictly the model follows your prompt.
What the AI does
Higher values follow the prompt more literally; very high values can look harsh or oversaturated; low values are freer.
When to use it
Raise it when the prompt is ignored, lower it when images look forced.
Where it works
In most Stable Diffusion apps (for example AUTOMATIC1111 or ComfyUI) this is a setting next to the prompt, not something typed in it. ChatGPT, Gemini and AI Compass have no such control.
Common mistakes
Maxing it out, which often makes images worse.
Changing CFG and steps together and not knowing which helped.
Example
Explain the CFG scale in Stable Diffusion with a simple analogy, and suggest a starting value for portraits and for landscapes.
Sampler
Also written: sampling method, scheduler
The method the model uses to remove noise step by step.
What the AI does
Different samplers give slightly different looks and need different numbers of steps.
When to use it
Try another sampler when results look noisy or too smooth.
Where it works
In most Stable Diffusion apps (for example AUTOMATIC1111 or ComfyUI) this is a setting next to the prompt, not something typed in it. ChatGPT, Gemini and AI Compass have no such control.
Common mistakes
Switching samplers constantly instead of fixing the prompt first.
Example
In simple terms, what is a sampler in Stable Diffusion, and how should a beginner choose one?
img2img
Also written: image-to-image, denoising strength, init image
Creating a new image from an existing picture plus a prompt.
What the AI does
Keeps the layout of your picture and repaints it; denoising strength sets how much changes (low = close to the original, high = more new).
When to use it
To restyle a sketch or photo while keeping its composition.
Where it works
In most Stable Diffusion apps (for example AUTOMATIC1111 or ComfyUI) this is a setting next to the prompt, not something typed in it. ChatGPT, Gemini and AI Compass have no such control. To change a photo in AI Compass, upload it in chat and describe the change.
Common mistakes
Using a very high denoising strength and losing the original layout.
Using images you don't have rights to.
Example
Explain how to turn my pencil sketch of a house into a watercolor painting with Stable Diffusion img2img, and which denoising strength to start with.
(word:1.3)
Also written: prompt weighting, attention weights, emphasis
Syntax that makes a word count more or less in a Stable Diffusion prompt.
What the AI does
(word:1.3) gives the word more influence; values below 1 give it less.
When to use it
When one part of the prompt keeps being ignored.
Where it works
Supported in most Stable Diffusion apps (exact syntax can differ between apps). Midjourney uses :: weights. ChatGPT, Gemini and AI Compass have no weights; put the important part first.
Common mistakes
Using extreme weights such as 3.0, which break the image.
Copying this syntax into other tools.
Example
Write a Stable Diffusion prompt for a garden scene where the red umbrella is emphasized with (red umbrella:1.3), and explain the syntax.
Every guide and dictionary card is also in the AI Compass app, with saved items, search and “Try it” buttons that open chat, image or video with the example filled in. Reading is free; creating uses credits, and the price is shown before you start.