To create an AI image, choose the final shape and purpose, describe one clear scene, generate a small first sample, and revise the most important mismatch. You do not need a long prompt full of quality adjectives. You need a result you can compare with a specific brief.
This guide uses Adobe Firefly’s interface as a concrete route and explains how to adapt the method to other tools.
1. Decide what the finished image must do
For this exercise, make a landscape image for the top of a personal page. The subject is a blue ceramic mug on a wooden desk. Keep it on the left so that a heading can sit on the right. The three pass conditions are simple: one intact mug, clear empty space for text, and a crop that keeps the whole subject visible.
Choose the aspect ratio before generating. A centered square composition can be difficult to convert into a wide header without cutting something off. If you need exact wording in the final design, plan to add it as editable text after the picture is settled.
2. Open an image-generation workflow and check its settings
In Firefly, Adobe documents the route as Image, then Generate image. Choose a model under General settings, set the aspect ratio, and select an appropriate content type when that control is available. Settings vary by model, so follow the options shown for your selection. Adobe’s generation guide describes the current route.
Check the displayed generation cost and available balance before submitting. Start with the smallest practical attempt. A preview or practice output does not need the most expensive size setting, and spending more does not guarantee that the composition will be right.
3. Write a prompt you can evaluate
Practice prompt: A single blue ceramic mug on the left side of a pale wooden desk, eye-level view, soft window light from the left, plain cream wall, wide landscape composition with clear empty space on the right, no text or logos.
Each phrase has a job: the subject, location, camera view, lighting, background and layout. If a requested detail does not matter to the deliverable, leave it out for the first attempt. Too many competing details make it harder to decide what to correct.
Save the exact prompt and model name. If the service offers automatic prompt enhancement, record whether you used it and preserve the resulting prompt when visible. Otherwise you may be comparing a different instruction from the one you intended.
4. Inspect the first image before changing anything
| What to check | A concrete failure | What to change next |
|---|---|---|
| Subject | An extra mug or a broken-looking handle | Simplify the subject description or use a local edit if available |
| Layout | The mug fills the space reserved for the heading | Strengthen the placement instruction or supply a layout reference |
| Lighting | The scene is too dark for the page | Change the lighting description while retaining the layout |
| Crop | The mug is cut off at the edge | Adjust framing or aspect ratio before polishing details |
| Text | Invented marks resemble a logo | Remove the marks and add any intended lettering separately |
Pick the failure with the greatest effect on the final use. Do not change the model, prompt, reference, crop and style in one attempt. You will lose the ability to tell which change helped. Save rejected versions too; they explain why you selected the final result.
5. Use the right kind of reference
A layout problem calls for structural guidance. In Firefly’s composition-reference workflow, open Composition, upload a reference and adjust its strength. A simple image you own showing where the mug and empty space belong can communicate arrangement more directly than additional adjectives.
A style problem calls for different guidance. Firefly’s Styles reference controls use a separate reference and strength setting. Use an image you have permission to upload, and decide whether you are borrowing its palette, lighting or overall treatment. A style reference does not establish exact subject preservation.
In Midjourney, Image Prompts guide a new creation from reference imagery. The documented prompt should describe the wanted final scene. If you need a precise change to an existing file, choose an editing workflow that supports that change rather than expecting a reference to copy the source exactly.
6. Make the second version answer one specific correction
If the mug is too central, keep the subject and lighting phrases, then replace the layout instruction with: “The mug occupies the left third; the right two-thirds remain an empty cream wall and desk.” If the composition is already correct, leave it alone and change only the defect that remains.
Compare versions at the size where the image will be used. A tiny texture improvement may disappear on a phone, while a poor crop becomes more obvious. Stop when the image meets the brief. More variations can produce more choices without producing a better deliverable.
7. Add text and verify the download
For a heading, use a text layer in your design editor instead of asking the generator to redraw the entire image around a word change. Ideogram’s Canvas Text Tool is one documented example of placing and formatting text separately within an image workflow.
Download the selected version and open the file. Check dimensions, format, unwanted marks, subject defects and the actual crop. Confirm the current usage terms for the plan under which you generated it. Keep the final prompt and reference alongside the file so you can make a related version later.
If you want a character instead of a general image
A reusable companion character begins with identity and personality decisions as well as appearance. Cherrypop’s character creator is the task-specific route for that. It is free to start with limits, and some media features require Premium or Cherries. A general image prompt alone does not create a complete companion.