Image guide

Make a consistent image set

Carry a subject and visual direction across images with references and traits.

Consistency starts from one approved asset, not from repeatedly rebuilding its prompt. Keep the subject, visual direction and decisions on the canvas where the next recipe can point to them.

Establish the source

Generate or upload one clear image. If the important traits are visual, run Describe and edit the resulting traits text until it names only what should persist: subject, silhouette, materials, palette, setting or treatment.

Describe costs 2 credits. It records text on the asset; it does not judge quality or change media.

Make the next images

Use the approved asset as a reference and choose a model whose live catalog entry accepts the needed image slot. Put persistent traits in the prompt, then state the one thing that changes: pose, view, expression, prop or scene.

Use count for a deliberate batch. Each result becomes its own asset with its own recipe and charge. Inspect the cards side by side and star the keepers.

For a near miss, choose Use current model, then change one variable so the reason for the new result stays legible. Use recorded model only when you want the fixed recorded inputs and provider model; it fails if that route is no longer configured.

Hand off the set

Name assets by their role, not by an attempt number. Add short notes only where the visual result cannot explain a decision. Export the starred assets' metadata when an agent needs to build the final deliverable, and download the ready media separately.

Before spending, read the model catalog and quote the request.