Negative prompt library for image models
A negative prompt tells a diffusion model what to avoid — extra fingers, blurry edges, watermarks, the wrong art style. The trouble is everyone copies the same giant block of tokens from a forum without knowing which ones matter for their model. This tool gives you a categorized, toggleable library so you build a focused negative prompt from the failure modes you actually care about, with guidance on how much each model needs.
How it works
Tokens are grouped into categories: anatomy errors (extra limbs, malformed hands), quality (low resolution, blurry, jpeg artifacts), artifacts (watermark, signature, text), and style avoidances (cartoon, oversaturated). You pick your model, toggle the tokens you want, and the tool assembles a deduplicated, comma-joined negative prompt. Because model needs differ, the tool adjusts its defaults — a rich set for Stable Diffusion 1.5, a trimmed set for SDXL, and a minimal set for Flux — and warns when you are likely over-doing it.
Tips and notes
- Match the negatives to the image. Anatomy tokens matter for people and animals; they are noise for a landscape. Toggle by relevance, not habit.
- Less is more on newer models. SDXL and especially Flux degrade if you stack too many negatives — start minimal and add only when you see a specific defect.
- Keep your own base. Paste a base negative prompt you trust and layer categories on top; the tool dedupes so you never double a token.
- Pair with the positive prompt builder. Generate your subject prompt first, then append the negative prompt here for a complete generation setup.