Is a negative prompt mandatory?

No. SDXL and SD 3 produce good images without one. It becomes useful for portraits, hands and photorealism, and for killing a recurring defect such as watermarks or text.

Can I put 'nsfw' in the negative prompt?

Yes, and it is a common way to steer SD 1.5 fine-tunes away from adult content, without any guarantee. Adult content and violence are not allowed anywhere on this site, whatever the negative prompt says.

Does the order of words matter in the negative prompt?

Probably a little: the first terms weigh more, as in the main prompt. Put the defect you care most about first.

Negative prompt in Stable Diffusion: what it is, examples

A negative prompt is a second text field where you list what the image must not contain: blur, extra fingers, watermarks, text, ugly colours. Stable Diffusion steers the generation away from those concepts at every step, which often removes a defect that adding words to the main prompt would not.

How it works

At each denoising step the model computes two predictions: one guided by your prompt, one guided by the negative prompt. The step starts from the second prediction and moves towards the first, by the difference between the two multiplied by the CFG scale (the guidance setting). So the negative prompt is not a filter applied at the end: it shapes the image from the first step. That is also why a negative prompt has no effect unless the CFG is above 1.

Negative prompts that work

General quality (SD 1.5, SDXL):

blurry, low quality, lowres, jpeg artifacts, watermark, text, signature, cropped, out of frame

People and portraits:

deformed, bad anatomy, extra fingers, missing fingers, extra limbs, mutated hands, poorly drawn face, asymmetric eyes

Photography:

painting, drawing, illustration, cartoon, 3d render, cgi, oversaturated, overexposed

Anime and illustration:

photorealistic, photo, realistic skin, worst quality, normal quality, monochrome

Start with the general list and add the block that matches your goal. Ten to twenty terms are plenty; long copied lists mostly add noise.

Weights in the negative prompt

Weights work as in the main prompt: (blurry:1.4) pushes harder away from blur, (text:1.2) removes stray letters. Raise a weight only for a defect that keeps coming back; above 1.5 the model starts distorting other things to obey.

When it does nothing

CFG scale at 1: the negative prompt is ignored by construction. Use 5 to 8 with SD 1.5 and SDXL.
Turbo and LCM models: SDXL Turbo runs without guidance and ignores the negative prompt; an LCM LoRA runs at CFG 1 to 2 and barely reacts to it.
Stable Diffusion 3.5 Large Turbo also runs without guidance, so the negative prompt is ignored there too; the other SD 3 and 3.5 models do use it.
Concepts the model cannot separate: "hands" in the negative prompt tends to hide the hands rather than fix their anatomy. Repainting the hand with inpainting can.

On this site

The inpainting tool already applies a strong negative prompt behind the scenes, so you only describe what you want. The prompt examples show the negative prompt, when one was used, next to the main prompt; copy both to start from a known result, then change one term at a time with a fixed seed.

What is a negative prompt?