Leonardo AI Consistent Character: References and Elements

To build a Leonardo AI consistent character, lock one reference image of your persona, use it as a Character Reference on every generation at Mid or High strength, and train an Element, Leonardo's custom LoRA, on a curated set of that character's images. Together with a fixed prompt block, those two features keep the same face across new outfits, scenes and poses without running anything locally.

Checked October 2026 against Leonardo's image guidance documentation, its guide to training custom Elements, its pricing page and its Flow State page. The app interface changes often; feature names are more stable than menu positions.

Leonardo is one of the few browser tools that offers both main consistency methods: reference-image guidance and custom training. This walkthrough builds a persona with them in order and is honest about the limits. For how Leonardo compares with every other method, including local ComfyUI workflows, see our consistent character AI guide.

How a Leonardo AI consistent character is built

Before you start

The character sheet matters more than any setting. If you cannot describe the persona in a few precise lines, no reference image will hold it steady. Our guide to camera settings for AI photos helps you write the lens and light part of the fixed prompt block.

Leonardo AI Character Reference: step by step

Leonardo's documentation explains the strengths plainly: Low gives the greatest flexibility at the cost of resemblance, Mid reasonable flexibility with more resemblance, and High the best resemblance with less flexibility. The same page adds a caveat worth repeating: Character Reference is not intended as a face swap feature and does not produce a perfect replica every time. Expect to discard some outputs.

An example prompt structure

Split every prompt into two parts. The fixed part never changes for the life of the persona; the scene part changes every post. Keeping them separate is what lets you change outfits and settings without the look drifting.

FIXED (same every time)
[character name or trigger], [age range], [hair], [skin tone],
85mm portrait, f/1.8, soft natural light, warm film color,
natural skin texture, realistic photo

SCENE (changes per post)
sitting at a window table in a busy cafe, cream knit sweater,
holding a cup, morning light, candid expression

Write the fixed part once, test it with Character Reference at High until a plain portrait matches your master, then save it somewhere you will copy from every time. If you train an Element, put its trigger word in the fixed part. When results drift, change one element of the scene part at a time so you can see which word caused it.

Training an Element for stronger consistency

Character Reference works image by image. An Element works across everything: it is a custom LoRA trained on a dataset of your character. Leonardo's guide describes Element training on SDXL models at 1024 by 1024 resolution, from an uploaded dataset, with default or custom advanced settings.

• Build the dataset from your best matches. Use Character Reference at High to generate the character from several angles, with different expressions and lighting, and keep only images where the face clearly matches the master.
• Vary everything except the face. Different outfits, backgrounds and framing teach the Element that the face is the constant.
• Exclude anything off-model. One wrong face in the dataset teaches the wrong face.
• Train, then test with a neutral prompt. Generate a plain portrait with the Element and compare it to the master before using it for posts.

The principles are the same as training a LoRA anywhere, and our LoRA training guide for consistent faces covers dataset curation in more depth.

Leonardo AI free credits and plan limits

Leonardo's pricing page lists 150 Fast Tokens per day on the free plan (checked October 2026). Paid plans list monthly Fast Token allowances and allow top-up tokens. Token cost per image depends on the model, size and features, and image guidance or higher resolutions can cost more, so watch the cost shown before each generation.

In practice, the free plan is enough to explore a look, lock a reference and test Character Reference. Building a full dataset and producing posts every day usually needs a paid plan. Check the pricing page for which features, including Element training and private generations, are included at each tier, because these change.

Leonardo AI review: is it good for consistent characters?

This is an assessment from Leonardo's documented features rather than a benchmark. Leonardo's strength is that both main consistency methods sit in one browser tool with free daily tokens to learn on. Its limit is control: you work within its models and settings, and it cannot match a local workflow where you choose every node. Many persona creators start in a tool like Leonardo, then move production to ComfyUI once the character is proven. Our guide to running ComfyUI from a tablet covers one way to keep that workflow portable.

For the full system, from persona design to monetization and platform disclosure, our AI Influencers program walks through every step.

A weekly production routine once the character is locked

Consistency is a habit as much as a setting. Once the master reference and the Element exist, a simple routine keeps the persona stable for months:

1. Plan the week's scenes first. Write the five to ten scene prompts you need, each describing only setting, outfit and action.
2. Generate in one session. Use the same Element, the same master reference and the same fixed prompt block for every scene, so lighting and color match across the week.
3. Cull hard. Compare every candidate with the master at full size and keep only clear matches. A feed with fewer, consistent images beats a feed with more, slightly different faces.
4. Log what worked. Save the prompt, strength and seed next to each keeper, so a good result can be repeated next week.
5. Review monthly. Put the month's posts side by side. If the face is drifting, return to the master reference and tighten the dataset or strength.

This routine is also where token budgets are won or lost. Exploring at full resolution burns tokens quickly; generating drafts small, then upscaling or regenerating only keepers, stretches both free and paid allowances.

Disclosure for an AI persona

A consistent AI character that posts like a person still has to be presented honestly. Platforms increasingly require AI-generated content to be labelled, and audiences react badly to finding out later. Make the persona's AI nature clear in the bio, use each platform's AI label where it applies, and disclose paid partnerships the same way a human creator would. Our guide to Instagram's AI label rules covers the Instagram side.

Troubleshooting

• Face drifts in full-body shots. Small faces carry less detail. Generate closer framing, or raise Character Reference strength for those shots.
• Outputs look like the reference photo, not a new scene. Strength is too high for the change you want. Drop to Mid and describe the new scene more clearly.
• Element output is off-model. The dataset had mismatched faces. Rebuild it from stricter matches and retrain.
• Running out of tokens. Lower resolution for exploration, and save higher settings for keepers.

Leonardo AI consistent character: FAQ