Lesson Content

Guidance scale as a style control.

Guidance scale (CFG scale) controls prompt adherence – but at extreme values it also affects style character: low guidance scale (1-5) produces creative, often more artistic and expressive interpretations; medium (7-12) produces balanced prompt adherence with generative quality; high (15-20+) produces highly literal prompt adherence that can sometimes appear stiff or over-processed. For artistic, expressive output: lower guidance scale paired with strong style descriptors. For precise, controlled output: higher guidance scale.

Scheduler selection.

Leonardo AI offers multiple sampling schedulers – the algorithm that controls how the generation process proceeds through its denoising steps. Different schedulers produce subtly different aesthetic characteristics from the same prompt: some produce smoother gradients and soft transitions; others produce sharper detail and higher contrast. Verify currently available schedulers at leonardo.ai and test several on the same prompt to understand their aesthetic differences for your content type.

Step count and detail.

Higher step counts (e.g., 40-60 steps vs. default 20-30) generally produce more detailed, refined output at the cost of longer generation time and higher token consumption. For high-detail production images where fine texture and detail depth matter: higher step counts. For rapid exploration: default step counts. The step count where quality improvement plateaus varies by model – test to find the efficient threshold for your most-used models.

Seed for reproducibility.

The seed parameter specifies the starting random state for generation – the same seed with the same settings produces highly similar results. Useful for: reproducing a strong result for modification, comparing the effect of a single parameter change while holding all other variables constant (controlled comparison), and sharing reproducible generation approaches with collaborators.