Parameters

What is CFG scale (guidance scale)?

CFG scale, or guidance scale, controls how strongly the model is pushed toward your prompt and away from the negative prompt at each step.

CFG stands for classifier-free guidance. At every denoising step the model makes two predictions, one conditioned on your prompt and one unconditioned or conditioned on the negative prompt. CFG scale is the multiplier on the difference between them: it decides how far to exaggerate the direction that your prompt points in.

At a low value, around 1 to 4, the model is barely steered. Output is varied, often beautiful, and frequently has little to do with what you asked for. At a moderate value, roughly 6 to 9, you get the usual balance: the image follows the prompt while retaining enough freedom to look natural. This is where most tools default.

Push higher, past 15 or so, and the guidance starts to break things. Colours oversaturate, contrast blows out, edges get crunchy, and the image takes on a burnt, over-processed quality. The model is being shoved so hard toward the prompt that it leaves the region of images that look real. Higher is emphatically not better.

One thing worth knowing: Dreamify does not expose CFG scale directly. The generator gives you the negative prompt, scheduler, inference steps and seed, with guidance held at a sensible default. If you find output is not following your prompt closely enough, the lever here is a more specific prompt and a targeted negative prompt rather than a guidance number.

Worth remembering

Try it in the generator

The advanced panel exposes the negative prompt, scheduler, inference steps and seed. Twenty images free, no account needed.

Start generating

Frequently asked questions

Can I change CFG scale on Dreamify?
No. The advanced panel exposes the negative prompt, scheduler, inference steps and seed. Guidance is held at a sensible default, so prompt specificity is your main lever.
Why does a high CFG make images look burnt?
Because the guidance direction is being scaled far beyond where the model's predictions stay valid. The result drifts outside the space of natural-looking images, showing up as oversaturation and harsh contrast.
Is CFG scale the same as prompt strength?
Loosely, yes. Different interfaces label it guidance scale, CFG, or prompt strength, but they are all controlling the same multiplier.

Related terms

Back to the glossary, or apply it on a use case or an art style.