Unlike Midjourney, which interprets natural language, Stable Diffusion (especially SDXL and SD 1.5/2.1 models) relies heavily on mathematical syntax. If you don't assign prompt weights or specify a proper negative prompt list, your output will often contain visual glitches, distorted limbs, and low-quality textures. Here is how to master Stable Diffusion prompt math.
1. Understanding Prompt Weighting Syntax
Prompt weights tell Stable Diffusion which words are most important. The default weight of any word is 1.0. You can increase or decrease this using brackets:
- Parentheses ( ): Increases the attention weight of the enclosed word by a factor of 1.1.
(castle)= weight 1.1((castle))= weight 1.21
- Brackets [ ]: Decreases the attention weight of the enclosed word by a factor of 0.9.
[castle]= weight 0.9
- Numeric weights (Recommended): You can specify exact weights inside parentheses followed by a colon and the decimal value:
(castle:1.3)= increases importance by 30%.(foggy weather:0.85)= dampens fog effect by 15%.
2. The Ultimate Negative Prompt List
Negative prompts are words you want Stable Diffusion to actively avoid. A strong negative prompt is critical to prevent anatomically incorrect fingers, blurry faces, and bad lighting. Copy-paste this standard negative list for photo-realistic renders:
(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, extra fingers, mutated hands, bad proportions, lowres, bad quality, worst quality, artifact, signature, watermark, username3. Balancing Positive vs. Negative Weights
If you set your positive weights too high (e.g., above 1.5), the image will look "burned" or oversaturated (artifacts). Keep weights between 1.1 and 1.3 for best results. Always place the most critical subject at the very beginning of the prompt box, as Stable Diffusion assigns higher priority to the first 75 tokens.