Experiences
Build an image prompt around what you want to see
Begin with the character, the visible action and the frame. Then describe the clothing, place and light that make that moment specific. A focused picture request gives you a clear basis for judging the output without turning every attempt into a longer list of unrelated instructions.
On this page
Decide the subject and camera relationship first
Specify whether the picture is a close portrait, a waist-up view or a wider scene. Keep the pose possible within that frame. If the face is your priority, a distant figure in a busy landscape sends a different visual instruction. Describe one main activity and choose background details that support it.
The selected companion already identifies the character in the supported tool. Add the outfit and surroundings relevant to this image rather than using the request to retell an entire conversation. A story premise and an image composition answer different questions.
Portrait example: a repair bench by the window
This request establishes the frame, a simple hand position and one source of light. The uncluttered background makes it easier to assess whether the character, clothing and activity were followed. Replace the important details for your own scene rather than adding several different scenes together.
Show this adult character from the waist up at a small bicycle-repair bench. Use a navy work shirt, one hand resting beside a wheel and soft afternoon window light. Keep the tools behind the character out of focus and the expression relaxed.
Anime example: a message at the harbour
For an illustrated scene, state the style and preserve the same basic visual structure. A single object in the character's hand and a clear time of day establish an action without prescribing every detail in the world. After viewing the output, adjust the framing or object placement if that matters most.
Create an adult anime character standing beside a harbour noticeboard at dusk. Give the character a teal jacket and a folded letter held at waist height. Use a three-quarter view, a quiet pier in the background and warm light from a nearby lamp.
Revise the result, not an imagined technical parameter
Use natural-language visual instructions unless the selected tool offers a specific setting. Weighted tags, negative prompts, seed values and sampler names from another generator are not automatically accepted here. A copied parameter string can suggest control the interface does not provide.
Look at the actual picture before making another paid request. Identify the largest mismatch, remove conflicting wording and change that part. Increasing prompt length alone does not guarantee better detail, consistent identity or a fixed number of successful images.
Your questions
Should I keep extending a prompt until the picture is right?
Not simply for length. Remove instructions that conflict and retain the ones that determine the intended image. A focused revision lets you compare results more clearly than changing clothing, framing, setting and style all at once without knowing which instruction was missed.
Will another image tool's negative-prompt syntax work here?
Only use specialist fields or syntax when the chosen Candy AI tool actually exposes them. Different generators accept different controls. A familiar parameter name in a copied prompt does not establish that the feature exists or behaves the same way here.
What should I change after an unsuccessful result?
Pick the discrepancy that matters most: for example, the frame is too wide or the visible object is missing. Clarify that instruction and inspect the next output. Keep the request coherent and check the charge before spending on further attempts.