Stable Diffusion Prompt Generator

The key elements are: (1) Main subject - what you want to see, (2) Style - artistic style or medium, (3) Quality - keywords like '8k', 'highly detailed', 'masterpiece', (4) Lighting - type and mood, (5) Details - specific properties like colors, composition. The more detailed, the better the result.

Prompt weighting controls word importance: (word) increases weight by 1.1x, ((word)) by 1.21x (1.1 × 1.1). [word] decreases weight by 0.9x. Alternatively, use (word:1.5) for precise weighting. Example: (blue eyes:1.3) makes the blue eyes more prominent.

Negative prompts define what should NOT appear in the image. Common negative prompts include: 'ugly', 'blurry', 'poorly drawn', 'bad anatomy', 'watermark', 'text', 'signature'. They're essential for high-quality results without typical AI artifacts.

Sampling Steps (20-50) determine how many iterations the model runs. More steps = more detail, but slower. CFG Scale (7-15) controls how strictly the prompt is followed. Lower values (5-7) = more creative, higher values (12-15) = more prompt-accurate. For most cases, 30 steps and CFG 7 are optimal.

Recommended quality keywords: 'masterpiece', 'best quality', 'highly detailed', '8k uhd', 'sharp focus', 'professional', 'award winning'. These signal the model to generate high-quality outputs. Combine several for best results.

For consistent characters: (1) Describe the character in great detail in your first prompt, (2) Save the seed of the best result, (3) Use the same seed for additional images, (4) Keep the character description identical, (5) Use LoRA or embeddings for even better consistency.

Recommended samplers: 'DPM++ 2M Karras' (fast, high quality), 'Euler a' (creative, variable), 'DDIM' (deterministic, consistent). For beginners, 'DPM++ 2M Karras' with 20-30 steps is a good start. Experiment with different samplers for different effects.

Stable Diffusion responds better to technical details and specific descriptions. Key differences: (1) Prompt weighting with () and [], (2) Negative prompts are central, (3) No parameters like --chaos or --stylize, (4) More focus on quality keywords, (5) Longer, more detailed prompts often work better.