Leonardo AI Guide: Models, Prompts, Video & Tips

Leonardo AI in 2026: The Complete Handbook for Images, Video, Editing and 3D

Open Leonardo in 2026 and it no longer feels like a single image generator. It is closer to a creative workspace that happens to contain many different AI engines: image generation, natural-language editing, reference-driven design, fast visual exploration, video, upscaling, custom style training and even 3D. That breadth is Leonardo's biggest strength - and the reason the interface can feel confusing at first. This guide focuses on the parts that genuinely improve results, explained in plain English and organized around a practical workflow rather than a tour of every button.

Leonardo AI in 2026 is no longer just an image generator — it is a full creative platform for models, prompts, editing, video, and professional workflows.

If you want the broader foundation first, see Understanding Artificial Intelligence. This handbook assumes no technical background, but it focuses specifically on using Leonardo well.

Leonardo AI in 2026: More Studio Than Single Model

Leonardo started in Australia in December 2022 with a creator-first idea: give artists and production teams more control than the early generation of text-to-image tools allowed. It quickly grew beyond its original game-asset focus. Canva acquired Leonardo in July 2024 while keeping the product as a separate platform, and the relationship has since deepened: Leonardo technology now feeds into parts of Canva's ecosystem, while Leonardo continues to add its own models, third-party models and production tools.

That history explains why Leonardo feels different from tools built around one flagship model. Midjourney is closely associated with its own visual engine; ChatGPT Images is centered on OpenAI's image stack. Leonardo is closer to a multi-model studio: choose the engine that suits the job, then use Leonardo's references, editors, workflow tools and asset management to carry the project forward.

The practical rule Leonardo works best when you choose a model for the job, not for the hype. A photoreal product shot, a poster with clean typography, a character-preserving edit and a high-resolution infographic are different problems - and they do not need the same engine.

The 10-Minute Quick Start

If you are opening Leonardo for the first time, resist the urge to explore every setting. You can learn far more by finishing one small image from prompt to final export. A sensible first session looks like this:

1. Start with one clear idea. Use the home prompt bar and describe the visual you actually need. Let Auto choose the model while you are learning; model selection matters more once you can name the problem you are trying to fix.

2. Generate a few variations. Do not spend ten minutes polishing the first prompt. Treat the first batch like sketches: you are looking for composition, pose and mood.

3. Choose the best direction, not the most finished image. A frame with the right composition is usually worth keeping even if clothing, text or small details are wrong.

4. Edit instead of restarting. Use Inline Editor for targeted changes - background, clothing, text, objects or scene details - while telling the model what must stay unchanged.

5. Upscale at the end. High-resolution output is wasted if the composition is still moving.

6. Save what worked. Keep the winning prompt, references and useful settings. A repeatable recipe is more valuable than one lucky image.

Choosing the Right Model Without Overthinking It

Leonardo now offers a large mix of first-party and third-party models. The list changes quickly, so memorizing model names is less useful than learning what kind of problem each one solves. As of September 2026, the table below is a practical starting point rather than a permanent ranking.

Model / family	Best use	What it is good at	Watch out for
Lucid Origin	General creative work	Cinematic concept art, social visuals, mood boards, attractive first-pass images	Can over-detail scenes; less logical than reasoning-heavy editors
Lucid Realism	Photorealism	Portraits, products, materials, analog-looking light and texture	May converge on similar facial aesthetics; film grain may be unwanted for clean design
Phoenix 1.0	Prompt adherence	Stylized illustrations, game assets, layouts with specific elements	Less photorealistic than Lucid; can be rigid with unusual anatomy or impossible objects
Nano Banana family	Editing and logic	Object swaps, scene changes, consistent subjects, text-heavy visual tasks	Standard versions can look flatter; complex jobs may cost more or take longer
Ideogram 3.0	Typography	Posters, T-shirts, covers, logos and designs where readable text matters	Not the strongest cinematic photorealism
GPT Image 1.5	Complex editing / layouts	Spatial logic, UI mockups, dense mixed text-and-image compositions	Can look over-polished and may be slower
FLUX.2 Pro	Commercial realism	Product shots, exact colors, high-fidelity branded visuals	Longer text is still less reliable than typography specialists
FLUX Kontext / Max	Instruction-based editing	Precise changes to an existing image while preserving the rest	Repeated edits can slowly degrade an image
FLUX Dev	Custom Elements	Training a repeatable character, mascot or house style	Needs a good dataset; generic output can feel less natural than specialist models
FLUX Schnell	Fast ideation	Testing composition and prompt ideas cheaply and quickly	Not a final-quality engine

Beginner rule: start with Auto or Lucid Origin until you can identify what is missing. Then switch models for a reason - typography, realism, editing, speed, high-resolution output or custom training - not simply because another model is newer.

Different Leonardo AI models are designed for different outcomes — from photoreal portraits and product renders to typography, fantasy illustration, and advanced enhancement.

Prompting Leonardo: Clear Usually Beats Clever

A common beginner mistake is to treat prompting like spell-casting: pile on adjectives, camera jargon and style names until something impressive appears. That can work, but it makes failures hard to diagnose. A more reliable approach is to build the prompt in layers and put the visual idea that matters most near the beginning.

A practical prompt formula

[MAIN SUBJECT] + [ACTION / POSE] + [ENVIRONMENT] + [LIGHT] + [CAMERA / COMPOSITION] + [MATERIAL / STYLE] + [MOOD] + [IMPORTANT CONSTRAINTS]

Example - weak prompt:

futuristic astronaut, cinematic, ultra detailed, beautiful, 8k

Example - stronger prompt:

A lone astronaut repairing a solar array on the lunar surface, Earth low on the horizon, hard sunlight casting long black shadows, wide 24mm documentary composition, realistic suit materials with dust on the boots, restrained NASA-inspired industrial design, quiet and isolated mood.

The second prompt works better not because it is longer, but because almost every phrase controls something visible: subject, action, environment, light, framing, material, mood or composition.

Seven prompt habits that make Leonardo easier to control

1. Put the main idea first. If the product, character, storm or building is the point of the image, make it the first thing the model has to resolve.

2. Start simple, then add control. A shorter prompt makes failures easier to diagnose. Add detail only when you know what problem the detail is meant to solve.

3. Describe relationships, not just objects. "The red chair is behind the woman, partly hidden by the table" gives the model more usable spatial information than a list of nouns.

4. Use visual language. Lighting direction, material, distance, angle and framing usually change the result more than vague praise words such as "masterpiece" or "epic".

5. Do not make one generation solve five problems. If you need a new outfit, a new location, a different camera and a new expression, change them in stages so you can see which edit caused drift.

6. Put exact text in quotation marks. When the model supports typography, quote the words you want to see and keep the string reasonably short.

7. Know when to stop prompting and change models. If a model repeatedly fails at readable text or logical editing, another model is usually a better fix than fifty more prompt variations.

A useful editing phrase "Change X while maintaining the exact facial features, camera angle, composition and lighting." Leonardo's Omni editing guides explicitly recommend stating what must stay unchanged. This is one of the easiest ways to reduce unintended drift.

Reference Images: The Shortcut to Consistency

A text prompt can describe a type of person, object or style. A reference image is what you use when the exact identity matters. Leonardo's newer Omni workflows can combine multiple references, while legacy Image Guidance modes can target specific properties such as style, content, character, depth, edges, sketches or pose.

Three practical ways to keep a character consistent

Fastest: reuse one clean reference. For a small image set, start with a clear character image and describe the facial features, clothing and details you want preserved.

More control: combine references. Use one image for identity, another for environment and another for lighting or color. Tell the model what role each reference should play.

Most repeatable: train an Element. For a recurring mascot, character or house style, a custom Element can be worth the setup. A varied dataset - different angles, lighting and backgrounds - teaches the subject better than a folder of near-duplicates.

The limitation to remember: keeping the same face is easier than keeping the same person, outfit, accessories and tiny details across twenty images. For a long sequence, consistency comes from a workflow - reference, edit, lock successful frames, reuse them - not from a single switch.

Edit the Image You Have Instead of Starting Over

One of Leonardo's most useful changes is the shift from full regeneration toward natural-language editing. Open an image, choose Edit, and the Inline Editor can route the job through models such as Nano Banana, FLUX Kontext or GPT Image. In practice, this means you can fix the weak 10% of an image without throwing away the strong 90%.

Color correction: Change the jacket from red to matte black. Keep the same person, pose, lighting and background.

Background swap: Replace the office background with a rainy Tokyo street at night. Keep the exact subject position, camera angle and facial expression.

Product variation: Keep the bottle shape and label unchanged. Change only the background to pale stone with soft morning sunlight.

Typography fix: Replace "NXT HORIZON" with "NEXT HORIZON" while preserving the layout, font weight and lighting.

Wardrobe change: Change the clothes to a dark blue astronaut training suit. Preserve exact facial identity, hair, pose and expression.

Strong results in Leonardo AI rarely come from one perfect prompt — they come from prompting, testing, comparing drafts, and refining the best direction.

Flow State and Blueprints: Explore or Repeat

These two tools solve opposite problems. Flow State is useful when you are still searching for a visual direction. Blueprints are useful when you already know the job and want a repeatable recipe.

Flow State: explore before you commit

Flow State is built for rapid visual discovery. Enter a prompt, scroll through a stream of options, then use 'More Like This' to narrow the direction. It works well when the idea is clear but the art direction is not - a cover mood, a character silhouette, a product campaign look. Because each image still costs tokens, treat Flow State like a brainstorming session, not an infinite feed.

Blueprints: repeat a workflow that already works

Blueprints package models, prompts and settings into ready-made workflows. Leonardo launched them with more than 50 presets for tasks such as product photography and combined image-video creation. They are particularly useful for beginners because you answer a few practical questions instead of configuring every parameter yourself. Leonardo has also announced plans for user-built Blueprints, sharing and API access.

Which one should you use? Use Flow State when you are exploring. Use Blueprints when you know the job and want a repeatable process. Use the full Image Creation interface when you need precise model-level control.

Realtime Canvas, Canvas Editor and Inline Editor

The names are easy to mix up, but the tools encourage different ways of working. Realtime Canvas is for sketch-led ideation: rough shapes and color blocks influence the image almost immediately. Canvas Editor is a more deliberate workspace for masking, inpainting, outpainting and local control. Inline Editor is the fastest route when you can simply describe the change you want in words.

For everyday edits, start with Inline Editor. Use Realtime Canvas when composition is easier to sketch than explain. Reach for Canvas Editor when you need a precise mask, controlled outpainting or manual intervention that a natural-language instruction cannot reliably express.

Upscaling: Leave It Until the Image Is Finished

Upscaling is easy to waste money on. It cannot repair a weak composition, broken anatomy or misspelled text; it only gives the final image more resolution and, depending on the mode, more inferred detail. Leonardo's Pro Upscaler offers Precise and Creative modes and can optionally repair common AI artifacts. Precise is the safer choice for faces and product work. Creative can add texture and surface detail, but that extra invention is exactly why it should be used cautiously when fidelity matters.

Leonardo Video: Think in Shots, Not Entire Scenes

Video is now a substantial part of Leonardo. Rather than forcing creators into one engine, the platform can act as a gateway to several video models. Its 2026 catalog has included Veo 3.1, Kling 3.0 Turbo, Wan 2.6/2.7, Seedance 2.5, FLUX video models and Leonardo's own Motion family, with different strengths and costs depending on the model.

A reliable beginner video workflow:

1. Finish the still image first. A controlled starting frame gives the video model less freedom to redesign the subject.

2. Describe the movement, not the whole picture again. Tell Leonardo what the camera and subject do: slow dolly in, fabric moves in the wind, subject turns toward camera.

3. Prototype cheaply. Test the motion and prompt with faster or lower-cost settings before paying for a high-quality render.

4. Use start and end frames when the model supports them. They make transitions more intentional and can reduce random drift.

5. Treat audio as its own creative layer. Some models can generate sound or dialogue, but timing, intelligibility and unwanted artifacts still need human checking.

For a deeper look at Google's video-generation ecosystem, see Google Veo 3: Next-Gen AI Video Generation. Leonardo is useful here because it can place Veo beside competing models in one workflow rather than locking you into a single provider.

3D Generation: Reference Quality Matters More Than Prompt Flair

Leonardo added 3D generation in August 2026 using Hyper3D's Rodin V2. The workflow is straightforward: provide a clean reference image - ideally several views when available - and Leonardo produces a textured 3D mesh that can be previewed and exported as a .glb file for tools such as Blender, Unity or Unreal Engine. Generations start at 400 tokens, so this is a production feature rather than a casual free-tier experiment, but it meaningfully expands Leonardo beyond flat images.

The best 3D advice is unglamorous: give the model a better reference. A clean object on a simple background with visible contours tells the reconstruction system far more about shape than a dramatic shot with smoke, motion blur and hidden edges.

Leonardo AI is not only about generating images from scratch — its editing tools make it possible to refine, replace, expand, and reimagine visual content with precision.

Elements and Custom Training: When References Stop Being Enough

If you create the same mascot, character or visual language again and again, attaching reference images to every generation becomes inefficient. Leonardo's Elements are LoRA-style fine-tunes designed to make that identity more repeatable. They are currently centered mainly on FLUX Dev and selected legacy workflows.

For training, variety matters more than sheer volume:

Use multiple camera angles and distances.

Include different lighting conditions.

Vary backgrounds so the model learns the subject rather than memorizing one scene.

Avoid near-duplicate images that teach the same pose repeatedly.

For a character, keep identity stable while letting clothing and environment vary if you want flexibility later.

After training, test different Element strengths instead of assuming the default is optimal.

When not to train: if you only need a handful of images, a good reference-image workflow is usually faster. Train an Element when the same identity or visual language has to survive across many generations or projects.

Four Workflows That Are More Reliable Than Random Prompting

1. Product campaign workflow

Start with a clean product photograph or reference.

Use an editing-focused Omni model to place the product in several environments while preserving shape, logo and label.

Generate a few campaign directions in Flow State if the art direction is not fixed.

Use Ideogram or a text-strong model only when typography must be generated inside the image; otherwise add final copy in Canva or another layout tool.

Finish with Precise upscaling for print or large-format use.

2. Consistent character workflow

Create one clean "master" character portrait with neutral lighting.

Reuse it as the first identity reference and describe important traits in text as well.

Generate key poses and lock the best versions rather than chaining every new image from the previous one.

For a long series, train an Element once the design is stable.

Use local edits to change clothes, props or environment without rebuilding the face.

3. Poster or social graphic workflow

Generate the background art separately from the final copy.

Use Ideogram or Nano Banana when text must exist inside the generated art.

Keep exact text strings short and quoted.

If spelling is mission-critical, treat AI text as a draft and place final typography manually.

Export multiple aspect ratios rather than stretching one image for every platform.

4. Storyboard-to-video workflow

Use Flow State or a fast model to search for scene composition.

Turn selected frames into consistent keyframes with references.

Correct faces, props and text before animation.

Use a video model appropriate to the shot: realistic physics, fast motion, dialogue or camera control are different requirements.

Generate short shots, then edit them together. A sequence of controlled five-second shots is usually more reliable than asking one model to create an entire scene in one pass.

Pricing and Tokens: Plan the Expensive Steps

Leonardo uses tokens because different models and operations have very different compute costs. That means a 'generation' is not a fixed unit: a quick image test and a premium video render can consume very different amounts. The figures below are a September 22, 2026 snapshot, not permanent pricing.

Plan	Monthly price*	Fast tokens	Privacy	Best fit
Free	$0	150/day	Public by default	Learning and occasional experiments
Essential	$12	8,500/month	Private Mode available	Regular hobby use
Premium	$30	25,000/month	Private Mode available	Active creators; relaxed image generation on selected models
Ultimate	$60	60,000/month	Private Mode available	High-volume creators; selected relaxed video generation

*Monthly prices shown before tax. Annual billing may be discounted. Check Leonardo's live pricing page before purchasing, because plans and token allowances can change.

Relaxed Generation is the detail that changes how paid plans feel in practice. On eligible first-party models, Premium and Ultimate users can continue generating at lower priority after fast tokens run out. Many third-party models still consume tokens every time because Leonardo has to pay external providers to run them. In other words, a workflow built around premium video engines can burn through a token budget far faster than one built mainly around Leonardo's own image models.

Commercial Use, Ownership and Privacy: Read This Before Client Work

Leonardo's current terms allow commercial use, including on the free tier, but ownership, visibility and privacy are not identical across plans. Paid subscribers can use Private Mode, which is the more appropriate setup for confidential client work and unreleased assets. Free-tier generations are public under the platform's current rules, so creators should understand what that means before uploading sensitive material.

For client work, unreleased products, private photographs or brand assets, make privacy part of the workflow rather than an afterthought. Check the active plan, switch on Private Mode where available, and confirm the current terms before a high-stakes commercial project.

Copyright caution: commercial permission from Leonardo is not a blanket clearance of third-party rights. A generated image can still create problems if it reproduces protected branding, uses a real person without permission, closely copies copyrighted material or relies on source assets you were not entitled to upload.

12 Common Leonardo Mistakes

1. Using one model for everything. A strong general model can still be the wrong tool for typography, local editing or exact color.

2. Writing a novel instead of a prompt. Longer is not automatically clearer. Start with the subject and build outward only when the added detail changes something visible.

3. Regenerating instead of editing. When 90% of the image works, repair the 10% that does not.

4. Changing five things at once. Sequential edits make it easier to preserve identity and composition - and easier to undo the change that caused trouble.

5. Using weak references. A blurry, cropped or inconsistent source image gives the model less reliable information to preserve.

6. Expecting a seed to lock identity. Seeds help reproducibility, but the model, references and edit strategy matter more for character consistency.

7. Upscaling too early. Finish composition, anatomy and text before spending tokens on resolution.

8. Trusting generated typography without checking it. Even strong text models can introduce spelling, spacing or layout errors.

9. Ignoring visibility settings. Public generations may be visible or reusable through platform features, which matters for confidential work.

10. Burning premium tokens during ideation. Use Flow State, fast models or cheaper settings while you are still choosing the direction.

11. Training an Element before the design is stable. Do not teach the model a character or style you are still changing.

12. Treating the first AI output as finished design. The strongest work usually combines generation with editing, typography, layout and human selection.

20 Leonardo AI Tips Worth Saving

1. Use Auto model selection for the first pass, then switch manually once you know what the current model is failing at.

2. Generate the composition cheaply; spend quality tokens only after the composition works.

3. When using multiple references, explicitly say which image controls identity, which controls style and which controls environment.

4. For edits, state what must remain unchanged as clearly as what should change.

5. Use nouns instead of pronouns in complex edits: "the woman in the red coat" is clearer than "her".

6. If you need exact text, use quotation marks and keep the phrase short.

7. If a model keeps ignoring one element, move that element earlier in the prompt.

8. Use "keep exact camera angle and framing" when changing backgrounds.

9. For a recurring character, keep one clean master reference rather than using a chain of increasingly edited images.

10. For product work, preserve the product first and let the environment vary, not the other way around.

11. Use Flow State for art direction discovery; use normal Image Creation for controlled production.

12. Use Blueprints for repeatable jobs when you do not need to understand every underlying setting.

13. Use Realtime Canvas when the geometry in your head is easier to sketch than to describe.

14. Use FLUX Schnell or another fast option as a prompt test before moving to a high-fidelity model.

15. Use a text specialist for typography and a realism specialist for photography; do not force one model to do both if a two-stage workflow is easier.

16. If an image is already excellent, choose Precise upscaling; if it needs more texture, test Creative upscaling on a copy.

17. For video, animate a finished still rather than letting the video model redesign the subject from scratch.

18. For 3D generation, use clean reference views with visible contours and minimal background clutter.

19. Create Collections early for long projects. Finding the right reference later matters more than people expect.

20. Keep a tiny prompt library of proven structures for portraits, products, environments and edits instead of reinventing language every time.

Where Leonardo Seems to Be Heading

Leonardo's direction is becoming easier to read. The product is moving away from the idea of a single image generator and toward reusable creative systems. Blueprints already package complex workflows into simpler interfaces; user-built Blueprints have been announced as a future feature. 3D generation extends Leonardo into production assets. Canva's 2026 ProSuite strategy also links Leonardo more closely with tools such as Affinity, Cavalry and Flourish, suggesting a broader pipeline in which AI generation, professional editing, motion and data visualization can move between products.

That may matter more than winning a benchmark with one image model. Models change quickly. A more durable advantage is orchestration: helping a creator pick the right engine, give it useful references, edit the result, turn it into motion or 3D, upscale it, organize the assets and repeat the process without rebuilding the workflow from scratch.

Who Gets the Most Value From Leonardo?

Leonardo makes the most sense for people who want one visual workspace instead of a separate interface for every model. It is particularly useful for concept artists, marketers, game teams, social creators, product designers and small studios that move repeatedly between ideation, editing and production.

It is less compelling when the job is narrow and another tool already solves it extremely well. A photographer who only needs cleanup may prefer a dedicated editor. A filmmaker committed to one video model may prefer that model's native interface. Someone who generates two images a month probably does not need Leonardo's full depth.

The Skill That Matters Most Is Creative Direction

Leonardo can make a striking image from one sentence. That is useful, but it is no longer the interesting part. The platform becomes genuinely powerful when you treat it as a production system rather than a slot machine: explore quickly, choose deliberately, switch models for a reason, lock good references, edit instead of restarting, and spend expensive compute only after the idea is working.

The human contribution is still the part the interface cannot automate away: deciding what the work is for, which mistake matters, what feels off, what deserves another iteration and when the piece is finished. Leonardo can shorten the distance between an idea and a usable visual asset. It cannot decide which idea deserved the effort in the first place.

In 2026, Leonardo AI is expanding beyond still images into broader creative workflows that include video, 3D assets, and content production for multiple formats.

FAQ: Leonardo AI in 2026

Is Leonardo AI free?

Yes. Leonardo currently offers a free plan with a daily token allowance, although generations are public and some higher-cost models or advanced features are limited.

Which Leonardo AI model is best?

There is no universal best model. Lucid Origin is a strong general starting point; Lucid Realism is aimed at photorealism; Ideogram is useful when typography matters; Nano Banana and FLUX Kontext are strong for editing; and FLUX Dev is useful for custom Elements.

Can Leonardo AI keep the same character across images?

Yes, within limits. Character consistency is better treated as a workflow than a guarantee: use one clean master reference for small projects, combine references when needed, and consider a trained Element when the same identity must survive across a long series.

Can I use Leonardo AI commercially?

Leonardo's current terms allow commercial use, but ownership, visibility and privacy differ by plan and can change. For confidential professional work, a paid plan with private generation is the safer operational setup; always check the current terms for your use case.

Does Leonardo AI generate video?

Yes. Leonardo includes its own Motion tools and access to third-party video models. The exact capabilities depend on the model and can include image-to-video, start/end frames, camera control and, in some cases, generated audio.

Can Leonardo AI make 3D models?

Yes. Leonardo added image-to-3D generation in 2026, currently using Rodin V2, with downloadable .glb files that can be taken into tools such as Blender, Unity and Unreal Engine.

Is Leonardo better than Midjourney or ChatGPT Images?

It depends on the job. Leonardo's main advantage is breadth: generation, references, editing, video, upscaling, custom training and 3D can live in one workspace. A dedicated tool can still be better when you only need one narrow capability.

What is the best way to learn Leonardo?

Start with Auto model selection and simple prompts. Then learn references and Inline Editing. Only after those feel natural should you move into custom Elements, advanced Canvas workflows or expensive video and 3D generation.