Practical guide
A dependable editing prompt is easier to troubleshoot when each sentence has one job. Use the structure below instead of stacking unrelated style words.
Start with the exact outcome: cinematic portrait, color grade, social DP or couple composition.
Use neutral subject details and avoid names or private information that do not help the visual task.
Describe the background, light direction and shadow readability in concrete terms.
Use eye-level, head-and-shoulders, 50mm feel or 4:5 framing only where it supports the final platform.
List identity, age, hair, clothing, pose and objects that should not change.
Exclude likely failures such as distorted hands, duplicate features, fake text, logos and waxy skin.
Put the task first, then the permitted subject, scene, light, camera and preservation rules. Finish with output framing and negative constraints. This order gives each sentence one responsibility and keeps the highest-priority instruction near the beginning. It also makes the prompt scannable on mobile: users can remove one scene or camera sentence without accidentally deleting the identity rules.
Replace vague phrases such as ultra cinematic or professional quality with visible decisions. Name warm side light, readable shadow detail, a simple charcoal background, an eye-level head-and-shoulders crop or subtle film grain. Concrete language helps you compare the result with the request. Adjective stacks often repeat the same idea while leaving the tool to guess the composition and preservation boundaries.
A useful portrait instruction can say: edit one consenting adult portrait; preserve face identity, age, hairstyle and proportions; keep the source outfit; use warm side light and a simple background; frame at eye level in 4:5; avoid changed identity, waxy skin, distorted eyes, text and logos. The exact visual values may change, but each part has a clear function and can be tested independently.
Most editing tasks do not need a real name, address, account handle or other private information. Describe only the visible subject and the permitted transformation. The Builder keeps its text fields in the browser session and does not put subject details in the URL or analytics event properties. Apply the same discipline when you paste the prompt into any third-party tool, review that provider's own data terms, and remove metadata or context that is irrelevant to the visual task.
Quick answers
Long enough to state the task, visual direction, preservation rules and avoid list. Remove repeated adjectives that do not change the instruction.
No. Use it for concrete failure modes and prohibited changes, not a second copy of the desired result.
Try the structure
Build a balanced cinematic portrait without locking the subject to one gender, outfit or location. The prompt focuses on believable light, readable facial detail and a clean social crop.
Use this prompt when the person and scene should stay intact and only the color treatment changes. It describes highlight warmth, shadow depth and skin-tone protection without pretending to be an official preset file.
A small display picture needs a readable face more than a complicated scene. This prompt keeps the composition centered, the background simple and the expression confident without turning aggressive.
Two-person edits need separate identity anchors and a balanced composition. This workflow describes each consenting adult independently, then protects relative height, pose and natural expression.
