Write an image request around a visible moment

Choose the subject and viewpoint first

Identify the subject, its action and any important clothing or lighting. Include a location when the surroundings contribute to the scene. You can leave out descriptive categories that do not matter; a request need not become an inventory of every object that might be visible.

Settle on the framing before expanding the description. A close portrait calls for different visible details from a distant full-body view. Choose your intended composition and keep it consistent while you assess how later revisions affect the picture.

Portrait example: choosing a record

If the cardigan appears in the wrong colour, clarify the green clothing instruction while retaining the shop and waist-up view. Compare that feature in the next result first, rather than changing the entire scene before you can assess the revision.

Frame this character from the waist up in a vinyl shop, wearing a green cardigan and studying the records. Place one hand on an unmarked album sleeve. Light the scene through a morning window and keep the shelving behind the character simple.

Anime example: waiting beside the timetable

The waiting pose and tram stop belong to the same visible moment. To try a different mood, adjust the expression or lighting while retaining that activity. Keeping the rest stable makes the effect of your change easier to identify in the next picture.

Create a side view of an adult anime character waiting by a tram timetable. Dress the character in a yellow raincoat with a closed red umbrella in hand. Use bright, cloudy daylight, a plain street background and scattered puddles on the pavement.

Describe the scene without assuming specialist settings

Describe the image in plain language and use the settings offered in our image tool. Sampler values, seed settings, negative prompts and weighted syntax depend on the particular generator; do not assume that a parameter copied from another tool is supported in the request you are making.

Keep every requested scene within our content rules. Rewording prohibited material or disguising it through unusual spelling does not make it acceptable, and prompt revisions must not be used to evade a restriction.

Your questions

Is a longer description always a better request?

Length alone does not clarify a scene. Keep the details that establish the picture you want, remove conflicting directions and leave out additions that do not affect it. Extra wording can make the request less coherent.

Can I bring specialised prompt syntax from a different tool?

Check the fields offered by the Candy AI tool in use. A setting or syntax accepted elsewhere is not automatically supported here, so do not assume copied parameters will provide the same behaviour or controls.

Which part of a missed image should I revise first?

Identify the most important difference from your intended picture and clarify that instruction. Review the next output before adding other changes or spending tokens on more attempts; a focused revision is easier to evaluate.