Negative Prompting
Steer a generation away from what you do not want, without weakening what you do want.
On this page
Negative prompting is the practice of telling the model what to leave out. It is genuinely useful for recurring clutter — stray text, watermarks, extra characters — and genuinely counterproductive when overused.
Overview
There are two ways to exclude something. You can name it as unwanted ("no text, no watermark"), or you can describe the positive alternative ("a clean empty gradient sky"). The second is usually stronger, because models respond better to descriptions of what to draw than to descriptions of what to avoid.
Negatives work best on concrete, recurring nuisances. They work poorly on abstractions: "not boring" and "nothing ugly" give the model nothing to act on.
Why this matters
Predictable clutter costs credits. Removing it up front is cheaper than refining it away afterwards.
On a mousepad, unwanted elements matter more than on a screen. Stray shapes near the edges get cropped awkwardly and stray text prints permanently.
What is worth excluding
- Text, letters, captions, and watermarks — the most common source of ruined AI artwork.
- Extra people or characters when you asked for one.
- Frames, borders, and mockup edges that fight the mousepad's own edge.
- Signature marks and fake logos the model invents.
- Busy background clutter in the area where your mouse will actually sit.
cyberpunk street scene, no text, no people, no clutter, no bad anatomy, no watermark, nothing ugly
wide cyberpunk street at night, empty of people, clean neon signage without lettering, uncluttered wet asphalt in the foreground
The improved version turns each exclusion into a positive description, which the model can actually draw.
Step-by-step guide
Step 1: Generate once without negatives
Find out what the model actually adds before pre-empting problems that may not occur.
Step 2: Note the recurring intruders
Only exclude things that showed up in more than one generation.
Step 3: Rewrite each exclusion as a positive where you can
"Empty foreground" outperforms "no objects in the foreground".
Step 4: Keep the negative list short
Three to five exclusions is plenty. Long lists dilute the whole prompt.
Step 5: Regenerate and verify
Check that the exclusions worked without flattening the design.
Step 6: Refine rather than pile on more negatives
If one intruder persists, remove it in a targeted refinement.
Best practices
- Always exclude text unless text is deliberately part of the artwork.
- Prefer positive descriptions of empty space over negative statements about clutter.
- Exclude concrete objects, never abstract qualities.
- Reuse a short, proven negative list across a series instead of rewriting it each time.
Common mistakes
| Mistake | Why it happens | What to do instead |
|---|---|---|
| Long lists of negatives copied from forums | They add noise and often exclude things that were never a problem. | Keep a short list built from what you actually saw. |
| Excluding something central to the subject | The model gets contradictory instructions. | Rewrite the subject instead of negating parts of it. |
| Using negatives as quality control ("no bad anatomy") | Quality is not a thing the model can subtract. | Improve the positive description or change the model. |
Frequently asked questions
Is there a separate negative prompt field?
Exclusions can be written into your prompt in plain language. The Studio shows the fields available for the model you selected.
Why does the AI still add text after I excluded it?
Text is a strong prior in most image models. Combine the exclusion with a positive description like "clean unmarked surfaces".
Do negatives cost extra credits?
No. They are part of the prompt.
Can negatives make a design worse?
Yes — over-long negative lists reduce the weight of your actual subject.
Should I exclude "blurry" and "low quality"?
Rarely useful. If sharpness is the issue, see Blurry or Low-Quality Design.
Do all models honour negatives equally?
No. Behaviour varies by model — see When to Use Each Model.
Related articles
- Writing Better Prompts — A practical structure for prompts that produce usable mousepad artwork.
- How AI Understands Prompts — What actually happens to the words you type, and why small wording changes can move a generation a long way.
- Building Better Prompt Chains — Reach a complex design through a deliberate sequence of prompts instead of one enormous one.
- Prompt Structure — Build prompts from a repeatable structure instead of guessing.
- Common AI Design Mistakes — The mistakes that most often ruin an otherwise good design.
Related blog articles
- Prompt writing — Prompt breakdowns, before-and-after rewrites, and prompt patterns that hold up in print.
Need more help?
- Contact support — for orders, billing, and account questions.
- Join the community — ask other creators and share what you make.
- Chat for Ideas — brainstorm a direction before spending credits.
- Open the AI Studio — put any of this into practice right now.