10 Common AI Prompting Mistakes and How to Fix Them
Spot vague, conflicting and overloaded instructions before they waste time. Each mistake includes a better way to ask.
1. Asking for “something amazing”
“Amazing” does not specify a subject, audience or output. Give a concrete goal: “Create a friendly three-sentence introduction for first-time visitors,” or describe a visible scene. The model can then make choices within useful boundaries.
2. Hiding the important subject
A long list of decorations can bury the main person or object. Put the subject and its action first, then add setting and secondary details. If the subject disappears in the result, remove competing objects before adding more adjectives.
3. Giving contradictory directions
“Bright noon sun in a dark midnight street” asks for incompatible lighting unless you explain the intended surreal effect. Choose the priority and clarify the relationship: “Midnight street lit brightly by a single artificial floodlight.”
4. Cramming a film into one short clip
Five scenes and ten camera moves are unlikely to fit a brief video. Write one shot with a beginning, action and ending. If the story needs several locations, make several prompts and assemble the clips later.
5. Forgetting what a reference must preserve
“Use this image” leaves the model guessing whether it should copy the person, pose, palette or style. Assign the reference a role: “Keep the person and camera angle; change only the background.” Check that the product supports that reference input.
6. Using vague camera language
“Cinematic” does not tell the model where the camera is or what it does. Specify a wide establishing shot, low close-up, slow side track or static frame when it matters. Use one main movement at a time.
7. Treating every model the same
An image generator needs a final visual arrangement; an image editor needs a source and changes; a video model needs action over time. Check model-specific controls for aspect ratio, references and audio rather than copying a prompt unchanged everywhere.
8. Assuming exact text or identity
A prompt cannot guarantee every letter in a sign or perfectly identical faces across frames. Verify the output. For critical text, use an editing workflow or add the typography afterward; for identity, use supported references and inspect each result.
9. Adding generic “quality” tags repeatedly
Repeating “ultra quality, masterpiece, 8K” seldom clarifies the image. Replace filler with visible materials, lighting, composition and a meaningful style. Keep the prompt easy to revise.
10. Changing everything after one failure
If you rewrite subject, angle, style and light simultaneously, you cannot tell what helped. Identify the biggest problem, alter one related instruction, compare with the previous output and save the working version.
A conflicting prompt
Show a sunny midnight street with lots of action in one second.
A quiet street at midnight lit by one bright artificial floodlight. One pedestrian crosses the frame during a six-second static shot.
The light source and manageable action resolve conflicting and overloaded instructions.
Weak: “Epic cinematic masterpiece, perfect quality, do everything.”
Improved: “A weathered lighthouse on a rocky coast at blue hour, wide shot from the beach, one warm window against cool fog, restrained editorial illustration.”