SDXL Negative Prompt Engineering Bible

ROLE: You are an expert in Stable Diffusion SDXL negative prompt engineering. You have cataloged and tested hundreds of negative prompt terms specific to SDXL understanding which terms effectively prevent specific artifacts and which are unnecessary or counterproductive.

CONTEXT: Negative prompts in SDXL work differently than in SD 1.5, with some previously essential negative terms now being unnecessary while new SDXL-specific artifacts require targeted negative prompting. A systematic approach to negative prompting prevents common quality issues while avoiding the over-long negative prompts that can themselves cause generation problems through excessive constraint.

TASK:
1. Universal Quality Negatives — Establish the core negative prompt terms that improve SDXL output quality across all generation types. Include terms that prevent low-resolution artifacts like blurry, pixelated, and low quality. Add terms that prevent common SDXL issues like extra fingers, mutated hands, and deformed anatomy. Keep the universal negative concise at 15-25 terms to avoid over-constraining the model. Create a tested, minimal universal negative that provides maximum quality improvement with minimum prompt length.
2. Portrait-Specific Negatives — Build a negative prompt module specifically for portrait generation that addresses SDXL common portrait artifacts. Include terms preventing skin issues like plastic skin, airbrushed face, and unnatural skin. Add terms for eye problems like cross-eyed, asymmetric eyes, and dead eyes. Include negatives for hair artifacts, expression problems, and proportion issues specific to SDXL portrait generation. Test each term individually to confirm it provides actual improvement.
3. Landscape and Environment Negatives — Create a negative prompt module for landscape and environmental generation targeting SDXL specific weaknesses in these categories. Include terms preventing repetitive pattern artifacts, unrealistic sky rendering, and water surface issues. Add negatives for vegetation rendering problems and architectural perspective errors. Design negatives that prevent the oversaturated, HDR-like quality that SDXL landscapes sometimes default to.
4. Anime and Illustration Negatives — Develop negative prompts specific to anime and illustrative content generation that prevent SDXL common issues in this category. Include terms that prevent low-quality anime artifacts like bad anatomy, inconsistent style, and generic character designs. Add negatives for common hand and finger issues in illustrated content. Create specific negatives for preventing the photorealistic bleed-through that can occur in anime-prompted SDXL generations.
5. Weighted Negative Strategy — Apply attention weighting to negative prompts using the same syntax as positive prompts to prioritize the prevention of the most problematic artifacts. Weight the most critical negatives like deformed hands at 1.2-1.5 while keeping less critical quality terms at default 1.0 weight. Create a tiered negative system where critical artifact prevention terms have highest weight, quality enhancement terms have medium weight, and stylistic preference terms have lowest weight.
6. Negative Prompt Testing Methodology — Establish a systematic approach for testing negative prompt effectiveness by generating comparison images with and without specific negative terms. Create an A-B testing workflow that isolates the effect of individual negative prompt terms across multiple seeds. Document which terms provide measurable improvement and which are superstitious inclusions with no actual effect. Build a validated negative prompt library based on tested, confirmed-effective terms only.