Practical guide
How to write a prompt for an AI image
A prompt structure that stays clear, testable, and easy to improve.
A clear structure for AI image prompts: subject, scene, composition, light, style, constraints, and focused revisions.
Define the intended result
Start by naming the output, the source material you already have, and the details that must remain stable. A precise task makes model and mode selection much easier.
Check the selected model’s current input, format, resolution, duration, and control limits before launching. A prompt cannot add a capability that the model does not support.
- Define one primary goal for the generation.
- Describe observable visual or motion details.
- Preserve only the constraints that matter to the result.
- Compare revisions under the same conditions.
- Review rights, consent, and publishing context.
Use clear, compatible inputs
Start by naming the output, the source material you already have, and the details that must remain stable. A precise task makes model and mode selection much easier.
Check the selected model’s current input, format, resolution, duration, and control limits before launching. A prompt cannot add a capability that the model does not support.
Improve one variable at a time
Start by naming the output, the source material you already have, and the details that must remain stable. A precise task makes model and mode selection much easier.
Check the selected model’s current input, format, resolution, duration, and control limits before launching. A prompt cannot add a capability that the model does not support.
Review the complete output
Start by naming the output, the source material you already have, and the details that must remain stable. A precise task makes model and mode selection much easier.
Check the selected model’s current input, format, resolution, duration, and control limits before launching. A prompt cannot add a capability that the model does not support.
Topics
- AI creation
- AI creation
- AI creation
- AI creation
FAQ
What should I check first for how to write a prompt for an ai image?
Confirm the required output, source material, model mode, and any detail that must remain unchanged.
Why can the same instruction produce a different result?
Generative systems are probabilistic, and the selected model, settings, source files, and inference can all affect the output.
Should I use a reference image?
Use a reference when composition, pose, palette, or identity matters, and only when you have the right to process that image.
What should I review before publishing?
Review the complete image or video, source rights, consent, privacy, context, and the Pixly publishing rules.