Leonardo Elements and Image Guidance: A Practical Guide

Leonardo has two systems for steering a look without rewriting the prompt. Elements are style add-ons you switch on with an adjustable weight, and several can run at once. Image Guidance lets you upload reference images and tell Leonardo whether to borrow their style, their content, a character’s face or their structure. Used together they are how you keep a series consistent across dozens of images.

Elements: stackable style layers

An Element is a trained style add-on (a look such as a painterly finish, a toy-like render, a vintage photo feel, a particular illustration style) that you attach to a generation. Each Element has a weight slider. At a low weight it tints the result; at a high weight it dominates. You can combine several, and the combination is itself a look you can reuse.

• Start with one Element at a moderate weight. See what it does on its own before you add a second.
• Stack with a purpose. A medium Element (watercolour) plus a mood Element (soft film grain) is a sensible pair. Two competing medium Elements usually fight.
• Lower weights for photoreal work. Elements are strongest when they push a stylised look; on photographs, keep them subtle or skip them.
• Check model compatibility. Elements are trained for specific model families, and the picker only offers those that work with your chosen model. If an Element you want is unavailable, the model is the reason.

Write the prompt as though the Element were not there. Describe subject, action, setting and light; let the Element handle the finish. Doubling up ("watercolour illustration" in the prompt plus a watercolour Element at full weight) tends to over-cook the effect.

Image Guidance: four kinds of reference

Image Guidance is where you upload one or more images and choose what Leonardo should take from each. The types available depend on the model, and the exact labels have shifted over time, but the roles are stable:

1. Style Reference. Borrows colour, lighting, texture and overall mood from the image, not its subject. Upload a frame with the palette and finish you want and describe a completely different scene in the prompt.
2. Content Reference. Keeps the subject matter and rough layout of the image while letting the prompt change the treatment. Good for turning a photo into an illustration of the same thing.
3. Character Reference. Preserves a character’s face and identifying features so they can appear in new poses and settings. This is the one most people are looking for.
4. Structural guidance (edge, depth or pose). Locks the geometry: outlines, spatial depth or the skeleton of a figure. Use it when composition matters more than anything else, such as matching a storyboard frame or a client’s layout.

Each reference has its own strength control. A common mistake is pushing every strength to maximum. Style Reference usually wants to be moderate so the prompt still has room; Character Reference wants to be fairly high; structural guidance wants to be high only if you genuinely want that exact shape.

Combining references without a fight

You can run more than one reference type at once, and this is where series work gets efficient. A typical set-up for a children’s book spread is a Character Reference of the protagonist at high strength, a Style Reference of an approved illustration at moderate strength, and a pose guide at moderate strength to place the figure. The prompt then only describes the new scene and action.

If the results start to look muddy, drop one reference at a time to find the culprit. Most conflicts come from a Style Reference whose palette disagrees with an Element, or a structural reference fighting the aspect ratio. Fix the inputs before raising strengths. For the still-image fundamentals of keeping a character stable, the approach in consistent characters applies here too: fix the description, fix the reference, change one thing per run.

A consistent-character routine

1. Generate the character once in a neutral pose, plain background, even lighting, using Phoenix or your preferred model (see Phoenix and models). Pick the cleanest result and save it as your master reference.
2. Write a short, fixed description of the character (hair, eyes, outfit, one distinguishing detail) and paste it unchanged into every prompt.
3. Upload the master as a Character Reference at a fairly high strength.
4. Add your Style Reference or Elements for the look of the series, moderate weight.
5. Generate the new scene with a locked seed, then vary only the seed if the pose or expression needs another try.
6. When a new image is better than the master (clearer face, better lighting), promote it to master and note the date you did so.

Settings at a glance

Try it

With a Character Reference of your protagonist loaded and one illustration Element at moderate weight:

The same girl with the short red plaits and green dungarees, now crouched at the edge of a rock pool, pointing at a small crab, late afternoon sun, gentle sea breeze, wide shot with the headland behind her

Keep "the same girl with the short red plaits and green dungarees" word for word in every prompt of the series. Change only the action and setting, and if the face drifts, raise the Character Reference strength one step before touching anything else.