Experiences
Write an image prompt around a single scene
Think of a picture prompt as a description of one visible moment. Name the subject and situation first, then add the visual choices you will use to judge the result.
In this article
Choose the details that define the frame
Identify the character, what they are wearing, what they are doing and where the light comes from. Include the setting when it matters. You can leave out any category that adds nothing to your intended picture instead of filling the request with extra labels.
Settle on a viewpoint before adding detail. A face-filling portrait and a distant view of the whole figure ask for different compositions. Begin with one of them so later changes have a clear visual reference.
Portrait example: a pause at the record shop
If the cardigan comes out incorrectly, revise the clothing instruction and leave the record shop and framing in place. That gives you a focused change to inspect in the next result.
Show this character in a red cardigan browsing a record-store shelf, with one hand resting on a plain album sleeve. Frame the portrait from the waist up, with an attentive expression, clear morning window light and uncluttered shelves behind.
Anime example: waiting at the tram stop
The request gives the character a task and a place within a single frame. To change its mood, try a different expression or light first, keeping the main action stable while you assess the effect.
An adult anime-style character in a yellow raincoat waits beside a tram-stop timetable, holding a closed red umbrella. Use a side-on view, bright overcast daylight and a simple street background with a few puddles.
Use plain descriptions unless a control is offered
Do not treat these examples as instructions for seeds, weighting, negative prompts or sampler settings. They are scene descriptions. Only use specialist parameters when the selected Candy AI tool actually supplies the relevant field or control.
Keep each request within our content rules. Rewording or misspelling a blocked request to evade a restriction does not make the underlying content acceptable.
Questions and answers
How much detail does a picture request need?
Include what makes the scene recognizable to you, then remove conflicts and unnecessary additions. A longer description can add competing instructions without improving the picture.
Will settings copied from another image tool work here?
Use the inputs and options available in your Candy AI tool. Another generator's syntax does not establish support for the same parameters or behavior in this workflow.
What should I change after an unsatisfactory result?
Choose the mismatch you care about most and revise that part of the description. Inspect the next picture before changing other details or paying for more attempts.