Experiences
Turn your scene into a clear image prompt
Picture the moment you want to show before writing the request. Introduce the character and action, then describe the visual details that matter to you. Those details give you something specific to compare with the generated picture.
Topics in this article
Decide what belongs in the frame
Say who appears, what they are wearing and doing, and how the scene is lit. Add the location when it contributes to the picture. There is no need to fill every descriptive category if it adds nothing to your visual intention.
Choose the framing before accumulating detail. A full-figure view from a distance is a different composition from a close portrait. Start with the one you want, then use it as a reference for judging later changes.
Portrait prompt: browsing records
If the red cardigan is not represented as intended, revise that outfit instruction without changing the shop or waist-up framing. Examine the next picture for that difference before adjusting other parts of the scene.
In a record store, show this character looking attentively through the shelves in a red cardigan. One hand rests on a plain album sleeve. Use waist-up framing and clear morning window light, with uncluttered shelves in the background.
Anime prompt: a wait for the tram
The scene combines a location and activity in one frame. Explore a mood change through the character's expression or the light first, keeping the waiting action in place so you can see the effect of that adjustment.
Show an adult anime-style character beside a tram-stop timetable, waiting in a yellow raincoat and holding a closed red umbrella. Frame the scene side-on in bright overcast daylight, with a simple street behind and a few puddles.
Plain language and supported parameters
These prompts specify what a scene looks like, not seeds, weights, negative prompts or sampler values. Use ordinary descriptions unless your selected Candy AI tool offers the relevant specialist field or setting; the examples do not establish support for those parameters.
Every request must stay within our content rules. Changing spellings or phrasing to get around a restriction does not make prohibited material acceptable to request.
Your questions
Should I make my image prompt as long as possible?
Include the details that define the intended scene, not extra words for their own sake. Remove contradictions and unnecessary additions, since a longer request can introduce conflicting directions.
Are other generators' prompt parameters supported here?
Use only the fields and controls available in the Candy AI tool you have opened. Another generator accepting particular syntax does not mean the same settings or behavior are available here.
What is a useful first revision after a result misses the scene?
Focus on the detail that matters most and rewrite that instruction. Compare the next image before changing anything else or committing tokens to further generations.