The parameters

The Advanced Model Settings window lists the fields each model accepts, so no model shows all of these. Each field shows a short tip beside it. When to change a parameter shows where the window is.

Image parameters

Parameter	What it does	Too high looks like
Guidance Scale (also Guidance, Cfg Scale, Cfg)	How closely the image follows the prompt. Low is loose and often more natural. High is literal.	Harsh color and contrast, stiff poses.
Num Inference Steps (also Steps, Sample Steps)	How many passes the model makes. More steps add detail, time and cost.	Little change past 25 to 50 steps. You pay for detail nobody can see.
Scheduler	The method the model uses to refine noise into a picture. Some need fewer steps.	It changes the look more than the quality. Try a few and keep what suits the film.
Aspect Ratio	The shape of the image or video.	A shape far from what the model was trained on warps the framing.
Strength (also Denoise)	In image-to-image, how far to move from the input. 0 keeps it, 1 ignores it.	At 1, the reference image stops mattering.
Prompt Strength	How much weight the prompt gets against the input image.	The input picture gets lost.
Clip Skip	Skips the last layers of the text reader for a looser, more stylized result.	Leave it at the default unless a LoRA’s notes ask for 1 or 2.
Scale Factor (also Scale)	On upscale models, how many times bigger the output is.	Big factors invent detail. 2× stays close to the source.

Two more fields show on models that support them. Negative Prompt lists what to leave out, such as “blurry, watermark, text”. It applies to every picture made with that design. Seed fixes the random start so the same prompt gives the same picture. Leave it empty for a new result each time.

Video parameters

Duration is the clip length in seconds. Cost rises with it, and most models get worse past the length they were trained on. Fps (or Frames Per Second) sets how many frames each second holds. More frames look smoother and cost more.

Text parameters

These apply to the text model that writes descriptions when you press Describe.

Parameter	What it does
Temperature	How varied the writing is. Low stays on script. High tries more ideas.
Top P	Another way to limit how far the model strays. Lower is more predictable.
Max Output Tokens	The longest reply the model may write. Raise it if descriptions stop short.