Experiences
Write a picture prompt you can refine
Describe what should be visible in the image. Start with the character and the main scene, then add the few details that will make a result easy for you to evaluate.
Use a small set of visual choices
Combine a subject, an outfit, a pose or action, a location and a lighting choice. You do not need every category in every prompt. The aim is to state what matters, not to fit as many descriptors as possible into the box.
Choose a single composition. Asking for a close-up and a distant full-body view at the same time creates competing instructions. A simple starting image gives you a better basis for the next attempt.
Example: a casual portrait
If the sweater is the wrong color, change that part of the request first. Keeping the setting and pose stable makes it easier to see whether the adjustment helped.
A portrait of this character in a dark green sweater, seated beside a café window, looking toward the camera. Soft afternoon daylight, a simple background and a relaxed expression.
Example: an illustrated scene
This request defines an action and a setting without adding several unrelated events. For a different mood, alter the lighting or expression instead of replacing the whole scene at once.
This anime-style character standing at a small flower stall at dusk, wearing a blue jacket and holding one wrapped bouquet. Warm stall lighting, a quiet street behind the character and a three-quarter view.
Separate writing advice from actual controls
A prompt example does not establish that your generator has seeds, weighting syntax, a negative-prompt field or a particular sampler. Use the controls offered by the tool rather than pasting unsupported parameters into ordinary prose.
Every generation remains subject to our content rules. Trying alternate spellings to force a blocked request is not a way to make that request acceptable.
Questions and answers
Should I write a long image prompt?
Use enough detail to identify the scene, then remove contradictions and details that do not affect the image you want. More words are not automatically more useful.
Can I use technical prompt syntax from another generator?
Do not assume another tool's parameters have the same effect here. A field or control needs to be supported by the Candy AI workflow you are using.
How should I fix an image that looks wrong?
Identify the most important mismatch and revise that detail. Review the new result before adding more changes or submitting repeated generations.