Negative prompts you can copy for cleaner AI images

blurry, low quality, low resolution, jpeg artifacts, noise, out of focus, overexposed, underexposed, oversaturated, distorted, deformed, watermark, signature, text

A good default for Stable Diffusion, Leonardo and other tools with a negative prompt field. In Midjourney, pick the few items you need and add them after --no; DALL-E and Flux generally ignore negatives, so describe a clean, sharp image instead.

asymmetrical eyes, crossed eyes, lazy eye, deformed face, distorted features, extra eyes, misaligned pupils, double chin artifact, bad teeth, waxy skin, plastic skin, uncanny, cropped head, out of frame

Use with photoreal checkpoints in Stable Diffusion or Leonardo when faces come out warped. Midjourney users can try --no deformed face, extra eyes; for models without negatives, ask for a natural, symmetrical face with relaxed expression.

extra fingers, missing fingers, fused fingers, too many fingers, malformed hands, mutated hands, extra limbs, missing limbs, extra arms, extra legs, bad anatomy, bad proportions, twisted torso, disconnected limbs, long neck

Most effective in Stable Diffusion and Leonardo; pair it with a clear pose description or ControlNet. Newer models like Flux handle hands better and do not use negatives, so describe the hand pose positively there.

cartoon, anime, illustration, painting, drawing, sketch, 3d render, cgi, digital art, airbrushed, plastic, doll-like, overly smooth, unrealistic lighting, oversharpened, hdr halo

Use when a photoreal prompt still drifts toward art styles in Stable Diffusion, Leonardo or OpenArt. In Midjourney combine --style raw with a short --no list such as --no illustration, cgi.

photorealistic, realistic skin, 3d, lowres, bad anatomy, extra limbs, messy lines, sketchy, blurry lineart, inconsistent eyes, off-model, muddy colors, jpeg artifacts, watermark, signature

Designed for anime checkpoints in Stable Diffusion and for Leonardo illustration models. Midjourney's Niji model rarely needs this; add --no realistic, 3d only if results drift toward realism.

cluttered background, distracting props, fingerprints, dust, scratches, smudges, warped label, misspelled text, distorted logo, uneven lighting, harsh reflections, color cast, low resolution, blurry, cropped product

Use in Stable Diffusion, Leonardo or OpenArt for e-commerce and ad visuals. Midjourney users can add --no clutter, text, props; for DALL-E or Flux, describe a seamless clean backdrop and spotless surface instead.

distorted perspective, bent walls, crooked lines, warped windows, impossible geometry, floating objects, cluttered, people, cars, construction debris, low resolution, blurry, overexposed windows, fisheye distortion

Useful for architectural visualization in Stable Diffusion and Leonardo, especially with wide-angle prompts. Remove people and cars from the list if you want a lived-in scene; in Midjourney try --no people, cars, distortion.

people, buildings, power lines, roads, cars, trash, oversaturated, fake sky, unnatural colors, repeated patterns, tiling, blurry, lowres, haze artifacts, watermark

Works in any tool with a negative field. Delete the man-made items if they belong in your scene. Midjourney users can append --no people, buildings, power lines to keep vistas pristine.

misspelled text, garbled letters, extra text, random words, gibberish, distorted typography, blurry text, cluttered layout, gradients, photorealistic, 3d, drop shadows, watermark, low resolution

Best used in Ideogram's negative prompt field or with SD-based models. Ideogram and Flux already render text well, so focus on clear quoted text first; in Midjourney, --no extra text can help on minimal designs.

unappetizing, burnt, moldy, greasy mess, plastic-looking food, artificial colors, oversaturated, messy plate, dirty utensils, cluttered table, harsh flash, blue color cast, blurry, lowres

Use in Stable Diffusion, Leonardo or OpenArt for menus and food ads. For DALL-E and Flux, describe fresh ingredients, steam and soft daylight instead of listing what to avoid.

scary, creepy, dark, horror, gore, violence, realistic photo, harsh shadows, sharp teeth, aggressive expression, messy lines, cluttered composition, text, watermark, extra limbs

Helpful in Stable Diffusion and Leonardo to keep characters gentle and consistent. Most platforms also apply their own safety filters; for models without negatives, describe soft colors, rounded shapes and friendly expressions.

low poly (unless intended), jagged edges, aliasing, noisy render, fireflies, flat lighting, plastic look, texture stretching, clipping geometry, floating parts, blurry, lowres, watermark

Use with SD and Leonardo 3D-style models; remove "low poly" if that is the style you want. In Midjourney, a short --no noise, flat lighting is usually enough.

blurry, anti-aliasing, smooth gradients, soft edges, photorealistic, 3d render, painterly, noise, jpeg artifacts, inconsistent pixel size, too many colors, watermark

Most useful with pixel-art checkpoints or LoRAs in Stable Diffusion. Outputs often still need downscaling to a true pixel grid; in Midjourney, --no blur, gradient, 3d can help keep edges crisp.

flickering, jitter, morphing, warping, melting faces, distorted hands, extra limbs, sudden cuts, inconsistent lighting, frame stutter, blurry motion, low resolution, watermark, text overlay

Use in video tools with a negative prompt field, such as Kling. Runway advises against negative phrasing, so describe a smooth, steady shot instead; simpler motion and fewer characters also reduce artifacts.

watermark, signature, logo, text, caption, subtitles, username, copyright mark, stamp, border, frame, ui elements, banner

A small add-on set to append to any other negative prompt in Stable Diffusion, Leonardo or OpenArt. In Midjourney, --no text, watermark, signature covers most cases; drop "text" if your design needs lettering.

airbrushed skin, plastic skin, waxy, overly smooth, beauty filter, porcelain skin, glossy skin, blurred skin texture, doll face, heavy makeup, oversharpened, hdr

Use for realistic portraits in Stable Diffusion and Leonardo, paired with positive cues like natural skin texture and visible pores. In Midjourney, use --style raw with --no airbrushed, plastic skin.