Generate images

Use an image node when you need campaign art, product exploration, a starting frame for video, or a controlled edit of an existing visual. This guide covers the full path from an empty node to a reviewed asset.

Outcome: a correctly sized image with a traceable brief, deliberate model choice, and enough review context to use it in a larger workflow.

Before you begin

Decide where the image will be used. Record the destination, aspect ratio, subject, message, and details that must not change. Gather approved product or brand references before opening the composer.

You can work without a reference for open-ended exploration. Use source media when the output must preserve a product, person, composition, material, or recognizable brand element.

1. Add an image node

Open a workspace and add a node from the canvas controls or canvas menu. Choose Image. The empty image node opens the media composer, where you can write the prompt, choose a model, attach compatible source media, set output controls, and review the current credit estimate.

Keep one clear purpose per node. If you need a hero image and three supporting crops, create separate nodes so each version retains its own prompt and settings.

2. Write a production brief

Treat the prompt as art direction, not a bag of keywords. Put the most important creative decision first, then describe what supports it.

Layer	Decide	Example
Job	What must the image accomplish?	Launch image for a vertical social story
Subject	What is present and what must stay recognizable?	One labeled bottle, front-facing, no extra packaging
Setting	Where does the scene happen?	Pale stone surface beside a sunlit window
Composition	How is the frame organized?	Product in the lower center, quiet upper third for copy
Treatment	What should the image feel like?	Warm editorial photography, natural texture, restrained contrast
Constraints	What must not appear or change?	No invented text, duplicate products, watermarks, or warped label

A complete first prompt could be:

Create a vertical launch image for a social story. Show one labeled bottle on a pale stone surface in warm morning light. Keep the bottle front-facing and recognizable, use an editorial photographic treatment, leave quiet space in the upper third for copy, and avoid extra packaging, invented text, or a watermark.

Use precise nouns and visible actions. “Soft window light from the left” is more actionable than “beautiful lighting.” If the image will receive typography later, describe the required negative space instead of asking the model to invent final campaign copy.

3. Choose generation or editing

Choose Generate when the prompt should define a new scene. Choose Edit when an existing image must remain the foundation of the result. The available choice depends on the selected model family.

Editing is appropriate for a controlled change such as replacing a background, adjusting the treatment, or creating a related variation. It does not guarantee pixel-identical preservation. Always inspect logos, labels, typography, faces, hands, materials, reflections, and small product features after an edit.

4. Choose the model

Open the model picker and remove any option that cannot meet the job's non-negotiables. Check whether the model supports generation or editing, the number of references you need, the required aspect ratio, and the required resolution. The picker can also show New, Popular, availability, and an estimate for the current route.

Use Choose a model for the job for the live capability comparison. A popular model is a useful starting point, not proof that it is the best route for every brand or visual style.

5. Add source media

Use the media control to upload a new source or choose a compatible asset already available in the workspace or brand media library. Attach only references that communicate a clear requirement.

• Use a product image to preserve shape, packaging, color, or label placement.
• Use a composition reference to communicate framing and visual hierarchy.
• Use a style reference to communicate palette, lighting, texture, or rendering treatment.
• Use several references only when the selected model accepts them and each one has a distinct role.

Conflicting references create ambiguous priorities. If identity matters more than style, say so in the prompt. The selected route controls the maximum reference count; the visible composer controls are the source of truth.

6. Set the output format

Choose the destination format before you spend time refining detail.

Destination	Common starting shape	Composition guidance
Story or Reel cover	9:16	Keep the subject away from interface overlays and leave vertical breathing room
Feed portrait	4:5 or 3:4	Reserve a clear hierarchy for product, headline, and supporting detail
Square post or tile	1:1	Avoid placing important information too close to the corners
Landscape banner	16:9	Decide which side should remain quiet for copy or interface elements

Aspect-ratio and resolution controls appear only when the model declares them. Higher resolution does not repair a weak composition. Lock the framing first, then choose the resolution needed for the delivery surface.

7. Generate a controlled proof

The first generation should answer one question: does this direction work? Keep the subject, references, aspect ratio, and main treatment stable. Change one major variable at a time so you can tell what improved the output.

A useful sequence is:

1. Test the core subject and composition at the lowest practical commitment.
2. Correct obvious prompt ambiguity before changing models.
3. Compare models with the same prompt and references when the direction is still uncertain.
4. Increase detail or resolution only after the composition is useful.

The estimate shown before generation is guidance for the selected route and settings. The final usage can differ when a workflow adds steps or the provider reports different billable work.

8. Review and refine

Review the result at full size and against the actual placement, not only as a canvas thumbnail.

• Message: can a viewer understand the intended idea without the prompt?
• Brand fit: do palette, lighting, texture, and mood match the active brand context?
• Fidelity: are products, people, logos, labels, and materials believable and consistent?
• Craft: check anatomy, perspective, edges, reflections, repeated objects, invented text, and visual artifacts.
• Layout: is required copy space genuinely usable at the final crop?
• Consistency: can this image sit beside the other campaign assets without feeling unrelated?

If the concept is wrong, revise the brief. If the concept is right but an element is wrong, edit from the strongest output or add a better reference. If the route lacks a required control, change models instead of repeatedly asking the prompt to override the model's capability.

9. Continue the workflow

Keep the strongest result as a reusable asset. An approved image can become a reference for another image, a starting frame for video, a visual input to a workflow, or campaign media prepared for a post.

Use References and output settings for deeper control, Generate video when the image should move, and Troubleshoot generation failures when a run is unavailable or fails.