Stable Diffusion negative prompts
Six focused sets you can copy, and a look inside the sampler at what a negative prompt really does.
| Set | Negative prompt | Copy |
|---|---|---|
| General quality A safe default for most SD 1.5 and SDXL images. | lowres, blurry, jpeg artifacts, worst quality, low quality, watermark, signature, text, cropped, out of frame | |
| Portraits and people Reduces the classic anatomy failures. | deformed, bad anatomy, extra fingers, missing fingers, fused fingers, extra limbs, disfigured face, cross-eyed, asymmetrical eyes, long neck, bad hands | |
| Photorealism Keeps a photo looking like a photo. | illustration, painting, drawing, cartoon, anime, 3d render, cgi, plastic skin, oversaturated, overexposed | |
| Anime and illustration Keeps a drawn style clean. | photorealistic, realistic, 3d, messy lines, sketch, muddy colors, extra digits, bad proportions | |
| Landscapes and architecture Stops people and clutter creeping in. | people, person, crowd, cars, power lines, tilted horizon, distorted perspective, hazy, washed out | |
| Product shots Clean commercial imagery. | cluttered background, dust, scratches, fingerprints, reflections of camera, logo, text, harsh shadows |
How a negative prompt steers the image
At every denoising step, Stable Diffusion makes two predictions and extrapolates from one toward the other: anchor + CFG × (positive − anchor). Without a negative prompt, the anchor is the empty prompt. With one, the anchor is your negative prompt. Drag the CFG slider and toggle the negative to see the step change (schematic, not to scale).
Balanced CFG, no negative: each step moves from the unconditional guess toward your prompt.
What this means in practice
- Negatives are directions, not filters. They change which way every step pushes, which is why a negative can also shift style, colour and composition, not just remove the listed thing.
- Only list problems you see. Each extra term blends into the anchor and weakens the others.
- CFG multiplies everything. A strong negative plus high CFG can over-correct. If images look harsh, lower CFG before adding more negatives.
- Weights work here too.
(extra fingers:1.3)strengthens a single negative term.
The sets above are built into the Stable Diffusion prompt generator, which merges and de-duplicates them for you. For Midjourney, use --no (see Midjourney parameters).
Questions
What is a negative prompt in Stable Diffusion?
A negative prompt is a second piece of text describing what you do not want. At every denoising step the sampler predicts noise for both prompts and pushes the image away from the negative prediction and toward the positive one, using the CFG scale to set how hard.
What is the best universal negative prompt?
A short general set works for most images: lowres, blurry, jpeg artifacts, worst quality, low quality, watermark, signature, text, cropped. Add category-specific terms (hands, anatomy, cartoon) only when you see those problems.
Do long negative prompts work better?
Usually not. Every term shares the same limited token budget, so a huge list dilutes each item and can flatten the image. Keep it focused on problems you actually see.
Does Flux or SD 3.5 need a negative prompt?
Less. Flux models are typically run without classic CFG, and many Flux front ends ignore negative prompts entirely. SD 3.5 supports them but needs fewer terms than SD 1.5 or SDXL.
Is --no in Midjourney the same thing?
It serves the same purpose. Midjourney’s --no parameter acts as a negative prompt for the listed words.