← Leonardo AI Creator Training Hub

Leonardo AI Omni Editor + CONTROL

A useful generation can be 90% right and still be unusable because of one wrong hand, object, face detail, background element, logo area or composition choice. The weak workflow throws it away. The CONTROL workflow protects the approved 90%.

Step 1: name the failure

• Object: wrong, missing, duplicated or malformed.
• Identity: character or product no longer matches the approved reference.
• Background: location, clutter or unwanted element is wrong.
• Composition: subject placement, crop or visual hierarchy fails the destination.
• Style: visual treatment drifted while subject/content is good.
• Text/design: typography or readable wording is the failed layer.

Step 2: preserve the source

Keep the original winning image. Edit a copy/version. Non-destructive thinking matters even when the interface makes iteration easy: you need a known-good reference point.

Step 3: write a local edit instruction

Describe what should change and, just as importantly, what should remain. Avoid rewriting the entire original prompt when only one component is wrong.

Step 4: compare against the source, not against imagination

Ask whether the edit fixed the identified failure without creating a more expensive failure elsewhere. Check face, hands, important props, brand symbols, text-safe space and edges around edited areas.

Step 5: escalate only when local editing cannot solve it

Regenerate the whole image when the underlying composition is wrong, the selected source never had the correct identity, or the visual concept itself has changed. Do not escalate because the first edit missed.

JR CONTROL ladder

1. Identify the failed component.
2. Change one relevant variable.
3. Compare to the preserved source.
4. Keep the correction only if the rest remains stable.
5. Repeat locally if necessary.
6. Regenerate globally only when the diagnosis is global.

Where Image Guidance helps

If local edits repeatedly damage identity or style, the real problem may be insufficient reference control. Move to Image Guidance + Elements rather than making the edit prompt longer.

Why this matters for teams and users

Teaching “regenerate until lucky” does not scale. Teaching diagnosis creates a repeatable support language: what failed, what must remain, what control should change and what evidence proves the revision improved the asset.

Return to Hub →