Prompt Strength

Also known as: prompt_strength

Prompt strength adjusts how much influence your text prompt has on the final image. Higher values push the model to follow the prompt more literally, while lower values let the model prioritize image coherence and natural aesthetics.

What It Does

This parameter acts as a dial between prompt faithfulness and image quality. Unlike guidance scale (which modifies the classifier-free guidance weighting), prompt strength is a separate model-level control that scales the text conditioning signal. At lower values, the model focuses on producing a visually coherent and aesthetically pleasing image, even if it departs from some prompt details.

At higher values, the model aggressively incorporates prompt elements, which can sometimes come at the cost of natural composition or visual harmony.

Adjusting It in CSF

Prompt strength is model-specific — not every model exposes it. To set it, edit a production design, open its Image tab, and click Advanced Settings to open the Advanced Model Settings modal. Pick the model you’re configuring; if it supports a prompt_strength parameter, it shows up in the list, usually as a slider from 0 to 1 with the current value beside it.

If you don’t see it for a given model, that model doesn’t offer this control — reach for guidance scale instead. Save stores your value for this design and model (only when changed from the default); Reset to Defaults restores it.

Value Ranges

Low (0.1–0.3)

Prompt is treated as a soft suggestion. The model prioritizes visual quality and may omit or reinterpret prompt details.

Mid (0.4–0.7)

Balanced prompt influence. Most prompt elements appear while maintaining natural-looking results.

High (0.8–1.0)

Strong prompt adherence. The model tries to include every detail specified. May produce less natural compositions in complex prompts.

Tips

Default is often 0.8–1.0 for most models. Only reduce it if the image looks unnatural.
If small prompt details keep getting ignored, try increasing prompt strength before adding more words.
This is model-specific — not all models expose this parameter, so it won’t always appear in the modal.
Works well in combination with guidance scale: increase guidance for concept adherence, prompt strength for detail inclusion.