Negative prompts tell the AI what to exclude from the image. They're essential for fixing common problems and achieving precise results.

What Are Negative Prompts?

A negative prompt is a separate field (in Stable Diffusion, Flux, and some Midjourney modes) where you list things you do NOT want in the image. The AI actively avoids these concepts.

Common Negative Prompt Terms

For general quality:

• "blurry, out of focus, low quality, low resolution, pixelated"
• "oversaturated, overexposed, underexposed"
• "artifacts, noise, grain (unless desired)"

For people/characters:

• "deformed, disfigured, bad anatomy, extra limbs, extra fingers"
• "crossed eyes, bad hands, mutated, ugly"
• "watermark, signature, text overlay"

For composition:

• "cropped, cut off, out of frame"
• "busy background, cluttered, chaotic"
• "border, frame, vignette"

Midjourney's Approach

Midjourney uses the --no parameter instead of a separate negative prompt field:

"A serene forest landscape --no people, buildings, roads"

Iterative Refinement Workflow

1. Generate first pass — Use your initial prompt to see what the AI produces
2. Identify issues — What's wrong? Extra elements? Wrong style? Bad composition?
3. Add negatives — Address specific issues with negative prompts
4. Adjust weights — Increase or decrease emphasis on specific terms
5. Vary seeds — Generate multiple variations to find the best starting point
6. Upscale and detail — Use img2img or upscaling for final polish

Prompt Weighting

Most platforms support weighting to emphasize or de-emphasize terms:

• Midjourney: "A forest:: sunset::2" (sunset gets 2x emphasis)
• Stable Diffusion: "(sunset:1.5)" or "((sunset))" for more emphasis
• DALL-E: Doesn't support explicit weights — use word repetition or positioning instead

Common Problems and Fixes

The 80/20 Rule

Focus 80% of your effort on the positive prompt and 20% on negatives. A well-written positive prompt needs fewer corrections than a vague prompt with extensive negatives.