Generation works best when it stays connected to layout. Treat every generated image as material for a canvas, not as the final artifact.
Pick a workflow
| Goal | Best starting point | What to compare |
|---|---|---|
| New direction | Prompt plus one or two references | Composition, subject position, and mood |
| Campaign variant | Duplicate a working frame | Copy length, crop, and color contrast |
| Product visual | Existing product asset plus generated background | Edge quality, scale, and text safety |
| Motion scene | Still frame with clean subject separation | Whether the subject can move or reveal clearly |
Generate with layout in mind
Before writing a prompt, decide where the image will live: hero background, product card, social post, storyboard frame, or motion scene. This gives the model a clearer role and makes the next editing step easier.
Write prompts around the role in the canvas. A useful prompt usually includes subject, composition, style, lighting, aspect ratio, and what should stay simple for text or motion.
Do not ask the model to solve layout and copy at the same time. Generate the visual material first, then compose text and supporting elements in the canvas where they remain editable.
Arrange before you polish
Place generated assets, references, and text blocks on the canvas early. Then check spacing, contrast, and reading order. If a generated image fights the layout, regenerate it with simpler composition instead of trying to fix every detail manually.
Use the canvas as the comparison surface:
- Place the strongest candidates near the target frame.
- Scale each candidate to the intended output size.
- Add real headline length before judging the image.
- Keep rejected candidates only when they explain a useful direction.
- Move temporary references outside the export frame before preview or export.
Remix from useful states
Duplicate a working frame before changing style, color, or motion. That gives you a visible history of directions and makes it easier to compare variants with a teammate.
Remix one axis at a time. Change color, crop, model, or motion separately so you can tell which change improved the result.
Keep reusable pieces
When a frame has strong hierarchy, save its structure as a repeatable starting point. Future projects can reuse the same layout while swapping copy, images, or animation timing.
Reusable pieces are usually structure, not pixels: frame ratio, text hierarchy, safe areas, motion order, and export settings. Keep those stable, then replace the generated media for each campaign.
Scale a batch safely
Before running many generations, make a small proof:
- One prompt.
- One target frame.
- One model or preset.
- Two or three outputs.
- One review export if the result will move.
Scale only after the prompt, crop, and text treatment work together. This avoids spending Flares on variations that repeat the same layout problem.