Guidance Scale
Definition
The Guidance Scale controls, in classifier-free guidance, how strongly a diffusion model follows the condition or prompt.
In classifier-free guidance, conditional and unconditional predictions are combined. The scaling factor changes the trade-off between prompt adherence, diversity, and possible artifacts.
Simply explained
Like turning a dial: higher means more prompt-oriented, but not automatically prettier or more correct.
Why it matters in practice
Relevant for image and other diffusion models when fine-tuning output style and prompt adherence.
Example from business
In image generation, multiple guidance values are compared to balance prompt fidelity and image quality for the specific model.
Limits and pitfalls
- Claiming a universal ideal value; applying Guidance Scale to every generative model; equating higher values with better quality.
Origin & history
Ho and Salimans described classifier-free guidance in 2022 as a method to combine conditional and unconditional score estimates.
Use in mid-sized companies
- Possible application: controlled variant generation in generative image production, provided the used model supports CFG.
Frequently asked questions
Is there an optimal value?
No, it depends on model, sampler, prompt, and goal. Do all image models use the same dial? No.
Sources
- Classifier-Free Diffusion Guidance · Ho & Salimans (2022), Classifier-Free Diffusion Guidance
Sources checked on 10.10.2026.
Editorially reviewed on: