Steps (Inference Steps)
Steps control how many denoising passes the model makes when generating. Higher steps = more detail, slower generation. For most use cases, the default (20–30 steps) is optimal. Use 50+ steps only for final outputs where maximum quality matters.
CFG Scale
CFG Scale controls how closely the model follows your prompt vs. generating freely. Lower = more creative/dreamlike. Higher = more literal/precise. Recommended range: 5–12. Values above 15 often produce over-saturated, unrealistic results.
Seed
The seed is a random number that determines the initial noise pattern. Same prompt + same seed = same output. Use a fixed seed when you want to iterate on a specific composition. Leave blank for random variation.
Negative Prompt
Words and phrases to explicitly exclude from the output. Useful defaults: "blurry, low quality, watermark, text, distorted hands, extra fingers, oversaturated"
→ Experiment with Advanced Settings