Guidance Scale is a parameter in AI image generation that controls how closely the model adheres to the text prompt versus allowing creative freedom in interpretation. Higher guidance values produce outputs that follow the prompt more literally and precisely, while lower values give the model more latitude to interpret and explore variations around the prompt's intent.

The parameter works by adjusting how strongly the model's generation process is steered toward the conditioning signal provided by the text prompt. At very high guidance scales, outputs tend to follow prompts precisely but may become oversaturated, overprocessed, or lack subtlety. At very low guidance scales, outputs may be more aesthetically organic but drift from the prompt's specific instructions. Finding the optimal guidance scale for a given prompt and desired outcome is a key skill in prompt engineering, and different models have different optimal ranges.

Understanding guidance scale helps creators diagnose why generations may not be matching their expectations and provides a technical control point for adjusting the balance between prompt fidelity and model creativity. For most use cases, moderate guidance values produce the best balance of prompt adherence and natural, aesthetically pleasing results.