Every AI video creator knows the feeling: the shot is almost perfect, except for the six-fingered hand, the flickering background, or the mysterious watermark. Negative prompts — telling the model what to exclude — are the direct fix. This guide covers how to write them well.
What negative prompts do
Most video tools accept two inputs: what you want (the prompt) and what you don’t (the negative prompt). The negative prompt steers the model away from its most common failure modes. Think of it as a quality filter: the main prompt describes the shot, the negative prompt bans the artifacts.
The universal artifact list
These are the defects AI video models produce most often. A version of this list belongs in nearly every negative prompt:
morphing faces, extra limbs, deformed hands, warping background, flickering, jitter, watermark, text overlay, logo, blurry, low quality
Keep it as a saved snippet and paste it in every time. It’s not glamorous, but it eliminates the most common reasons generations get discarded.
Writing targeted negatives
Beyond the universal list, write negatives for the specific shot:
- Crowd scenes: add merging faces, duplicate people, floating heads.
- Close-ups: add asymmetric eyes, distorted teeth, plastic skin.
- Product shots: add warped logo, misspelled text, deformed packaging.
- Dark scenes: add noise, grain overload, crushed blacks (some grain is fine; specify when it’s too much).
- Camera moves: add judder, stuttering motion, teleporting subject.
What NOT to put in negatives
- Style preferences. “Boring, ugly” are subjective and unpredictable — the model doesn’t share your taste. Negatives work on concrete artifacts, not aesthetics.
- Things already absent. Negating “cars” in a forest scene wastes tokens and can confuse the model into thinking about cars.
- Contradictions with the main prompt. If your prompt says “crowd cheering” and your negative says “people,” you’re fighting yourself.
- Overly long lists. Twenty negative terms dilute each other. Keep it under a dozen, focused on real risks for the shot.
Tool differences
Not every tool exposes negative prompts the same way. Some have a dedicated field; others expect them appended to the main prompt (often prefixed or in brackets — check your tool’s docs). A few newer models handle artifacts well enough that negatives matter less. Learn your tool’s mechanism once, then reuse the same lists everywhere.
Negatives vs. fixing in post
Negative prompts reduce artifacts; they don’t eliminate them. Budget for both: good negatives cut your discard rate significantly, and the editing workflow handles what slips through (trimming glitch frames, reframing problem areas). The combination is what produces clean final videos.
Build your library
Save three negative snippets: universal (the artifact list above, used always), people (face/hand/body terms, for character shots), and products (logo/text/packaging terms, for commercial work). Paste the relevant ones per shot. In a month, you’ll stop thinking about negatives entirely — they’ll just be part of your reusable prompt template.