In Cafa AI Images, you describe a picture and the image generator creates a new interpretation. The clearer you are about the subject and composition, the easier it is to get close to what you imagined.
Use Images when you want something new. Use Edit Image when you already have a picture and want to change part of it. In either tool, expect to try more than once and inspect the small details.
Start with a prompt structure
Describe the subject, action or state, setting, composition, lighting, color, and intended medium. Put essential elements early and avoid contradictory directions.
Define the subject and relationships
State the number of important subjects, what they are doing, and how they relate. ‘Two cyclists speaking beside their bikes’ is clearer than ‘people with bikes.’ For products or interfaces, generated text and exact geometry may be unreliable; plan to add precise typography in a design tool.
Control composition and viewpoint
- Close-up emphasizes detail; wide shot emphasizes environment.
- Eye-level feels neutral; low angle can feel imposing; overhead shows arrangement.
- Centered composition feels formal; off-center space can leave room for copy.
- Foreground, middle ground, and background cues create depth.
- Portrait, landscape, and square describe the intended frame, though exact controls depend on the tool.
Describe light and color concretely
Terms such as soft diffused daylight, hard side light, overcast exterior, warm practical lamps, or cool backlight communicate more than ‘beautiful lighting.’ Name a restrained palette when consistency matters. Avoid stacking incompatible lighting conditions.
Describe qualities instead of copying a creator
Specify medium, era, texture, shape language, contrast, and mood: for example, ‘flat editorial illustration with geometric forms and a limited earth-tone palette.’ This gives useful direction without depending on close imitation of a particular living artist’s identity.
Refine one variable at a time
Choose the closest result and identify the most important mismatch. Adjust composition, subject detail, light, or color while preserving successful elements. If every instruction changes at once, comparison becomes difficult.
- Pass one: subject and composition.
- Pass two: lighting and palette.
- Pass three: remove distracting details.
- Pass four: inspect edges, hands, faces, reflections, and embedded text.
When to use image editing
Use the editing workflow when you need an uploaded image to remain the reference. Describe the region or property to change and what must stay untouched. Results can still alter unintended details, so compare carefully with the source and retain the original file.
Limitations and responsible use
Generated images may contain distorted anatomy, impossible objects, unreadable text, unwanted stereotypes, or misleading details. Do not present synthetic documentary evidence as real. Consider consent and potential harm when depicting identifiable people, and verify that your use meets applicable platform, client, and rights requirements.