Leonardo AI is a Sydney-built, Canva-owned platform for generating and editing images and short video from text prompts, sketches, and reference images.

It has grown from a 2022 invite-only beta into one of the more feature-dense studios in the generative media category, built around a web-first workspace rather than a chat window or a Discord server.

Editorial Verdict

Leonardo AI earns its reputation as a creative workshop rather than a slot machine for images. Its combination of AI Canvas editing, custom model training, and access to a growing roster of first-party and third-party models under one token-metered subscription is hard to match at a comparable price.

Paid tiers grant full commercial ownership, and the free tier remains one of the more usable no-cost entry points in the category.

The trade-offs are real, though. Token consumption is uneven and can be hard to predict, Leonardo’s video tools remain a secondary feature next to dedicated platforms like Runway or Kling, and the platform’s raw image aesthetics now trail category leaders like Midjourney V8 and Flux 2 Pro on out-of-the-box polish.

Leonardo is best understood as a control-first platform for people who want to direct outputs precisely — game artists, character designers, and marketing teams building repeatable visual systems — rather than a tool for people who want the single most beautiful image from a one-line prompt.

What Is Leonardo AI?

Leonardo AI (styled Leonardo.Ai) is a generative AI platform for producing images, short video, and design assets from text, sketches, and reference material.

It positions itself less as a single model and more as a creative operating system: a shared workspace where Leonardo’s own models sit alongside licensed third-party models from providers like Google, Black Forest Labs, and Kuaishou.

The platform is aimed at a wide range of users — solo creators, game developers, marketers, and product teams — who need more granular control over output than a single-prompt tool typically offers.

Its core differentiators are the AI Canvas editing environment, a custom model training system (Elements), and a REST API that mirrors the web app’s parameters exactly, so a generation configured visually can be exported directly as runnable code.

Company Background: From Sydney Startup to Canva’s AI Engine

Leonardo AI was founded in December 2022 in Sydney, Australia, by CEO JJ Fiasson alongside co-founders including Jachin Bhasme, Chris Gillis, Peter Runham, and others, according to Wikipedia’s company profile.

The founding team drew on backgrounds in gaming and software engineering, and the product initially targeted game developers and artists who needed a way to train AI models on their own art styles rather than rely on generic outputs.

Early access was gated through Discord, and demand outpaced the invite queue quickly. By late 2023, the company raised roughly A$47 million (about US$31 million) in a Series A round led by Blackbird Ventures, with participation from Sierra Ventures, Smash Capital, and Samsung Next, among others.

Leonardo introduced Phoenix, its first in-house foundational image model, in June 2024, and by the end of that year the platform had surpassed 29 million registered users, per reporting from Forbes Australia.

In July 2024, Canva — Australia’s largest privately held software company — acquired Leonardo AI in a cash-and-stock deal, as detailed by Capital Brief’s coverage of the acquisition.

Financial terms were not disclosed. Leonardo continues to operate as a distinct product with its own subscription plans rather than folding directly into Canva’s pricing, while also powering select generation features inside Canva’s own Magic Studio tools.

The exact current headcount is not officially published; third-party trackers place it in the low hundreds as of mid-2026, up from roughly 120 at the time of the acquisition, but Aiverda Labs was unable to verify a precise figure from an official Leonardo or Canva source.

It’s worth noting that Leonardo’s pre-acquisition product had a rockier content-moderation history, including reports of users generating non-consensual explicit imagery before stricter filters were introduced — a pattern documented in Fortune’s reporting around the Canva deal.

Leonardo’s current help documentation describes two layers of moderation — prompt-level filtering and output-level NSFW flagging — though the company itself acknowledges that moderation is an ongoing, imperfect balancing act.

Core Features and Creative Tools

Image Generation Models: Phoenix, Lucid Origin, and Lucid Realism

Leonardo’s first-party model lineup centers on three families. Phoenix (now on version 2.0) is built for strict prompt adherence and multi-element scene composition, and it handles in-image text rendering better than most competitors, though it can be rigid about following literal instructions even when they’re biologically or physically implausible.

Lucid Origin is positioned as the aesthetic generalist — strong color, Full HD output, and reliable text rendering for graphic design work. Lucid Realism is tuned specifically for photorealism, simulating film grain, lens characteristics, and skin texture for portrait and product photography.

Independent comparisons published in 2026 generally place Leonardo’s own models a tier below the current photorealism leaders — Midjourney V8 and Black Forest Labs’ Flux 2 Pro — on raw out-of-the-box polish, while crediting Leonardo with stronger prompt adherence and far more granular control over the generation process.

AI Canvas and Real-Time Editing

The AI Canvas is Leonardo’s post-generation editing surface: inpainting, outpainting, background removal, and resolution upscaling all happen inside the same workspace used for generation, rather than requiring a separate tool.

A Real-Time Canvas mode lets users sketch a rough composition and watch the AI interpret it live, which is useful for creators who think spatially before they think verbally. Reviewers consistently point to this editing layer as a meaningful advantage over prompt-only tools like Midjourney or DALL-E 3, which offer little to no native post-generation editing.

Motion: Leonardo’s Video Generation Tools

Leonardo’s first-party video models — Motion 1.0, Motion 2.0, and Motion 2.0 Fast — animate still images or generate short clips from text, with camera controls for pan, zoom, and tilt. Leonardo also gives users access to licensed third-party video models, including Google’s Veo 3 and Veo 3.1, Kling 2.5 Turbo, and Sora 2, inside the same interface.

Motion is a genuinely useful feature for turning concept art into a short animated clip or a social media teaser, but it’s not positioned — or reviewed — as a primary video production tool.

Dedicated platforms like Runway and Kling generally produce longer, more coherent clips, and most comparative reviews describe Leonardo’s video capability as a solid secondary feature rather than a reason to choose the platform outright.

Custom Model Training and Character Consistency

Leonardo’s Elements system lets users train custom models on their own reference images — a character, an art style, a product line — and reuse that trained model across future generations.

This is frequently cited by professional users as the single biggest reason they choose Leonardo over prompt-only competitors, since it solves the consistency problem that plagues most one-shot image generators when producing a series of related assets (a game character across multiple poses, for instance).

3D Texture Generation

Leonardo also offers a 3D texture tool that generates albedo, normal, and roughness maps from a text prompt or an uploaded OBJ file, aimed squarely at game developers working in engines like Unity or Unreal who need to speed up asset texturing.

Multi-Model Access: Third-Party Integrations

Beyond its own models, Leonardo functions as an aggregator for a growing list of licensed third-party image and video models, including Veo 3/3.1, Flux.1 Kontext and Kontext Max, Ideogram 3.0, Kling, Sora 2, Seedream, and Google’s Nano Banana and Nano Banana Pro.

This is a meaningful differentiator: rather than committing to one vendor’s aesthetic, users can compare outputs from several leading models inside a single subscription and billing relationship.

The trade-off is that “unlimited relaxed generation,” Leonardo’s overflow feature for token-exhausted accounts, applies only to Leonardo’s own first-party models — every third-party model always consumes tokens at full rate, an important distinction that’s easy to miss when comparing plans.

Developer Experience and API

Leonardo AI provides a REST API at https://cloud.leonardo.ai/api/rest/v1, authenticated with a Bearer token, alongside official SDKs for Python and TypeScript, per Leonardo’s developer documentation.

The API mirrors the web app closely — a generation configured through the visual interface can be exported as ready-to-run API code via a “Get API Code” feature, which removes a lot of the usual trial-and-error in matching parameters between a prototype and production.

API access is billed separately from web app subscriptions. New developer accounts receive $5 in free API credit to test generations before committing to a paid plan, based on Leonardo’s own API pricing page.

Default limits allow 10 concurrent image generation jobs, with higher concurrency available through custom enterprise plans; Leonardo’s documentation frames rate limits, concurrency, and queue depth as three distinct concepts that developers need to plan around separately.

Webhooks are supported for asynchronous job notification, which is the officially recommended alternative to polling. Beyond the concurrency figures Leonardo publishes, granular per-minute request-rate limits are not fully detailed in public documentation — developers with high-volume needs are directed to contact Leonardo directly for a custom plan.

Security, Privacy, and Data Handling

Leonardo states it maintains SOC 2 certification with periodic external recertification audits, according to its own Data Processing Addendum. Data in transit is encrypted with TLS 1.2 or higher, and data at rest uses AES-256 encryption. The company also commissions an annual third-party penetration test and runs ongoing vulnerability scanning.

On privacy, Leonardo processes personal data under GDPR principles and offers a formal Data Processing Addendum for enterprise customers.

Content ownership and machine-learning usage differ sharply by plan: free-tier generations are public by default, and Leonardo retains broad rights to reproduce and distribute that public content, though free users still receive a non-exclusive commercial license to their own creations.

Pricing Structure and Token Economy

Leonardo prices its plans in tokens rather than a fixed number of images, with token cost varying by model, resolution, and quality settings — a single Lucid Origin image might cost roughly one token, while an Alchemy-enhanced render can cost eight to sixteen.

Paid tiers include a Rollover Bank that carries unused tokens into the next billing cycle (up to a plan-specific cap), and separately purchased Token Top-up packs never expire, which rewards bursty or project-based creators over those with steady monthly usage.

As of Aiverda Labs’ research, the live pricing structure on leonardo.ai/pricing is organized as Free, Essential, Premium, and Ultimate solo plans, plus Team and API tiers.

User Experience and Interface

Leonardo’s onboarding is more involved than single-prompt competitors: before generating, users select a model, set dimensions, choose a quality mode, and optionally apply style references or Elements. That extra setup is a deliberate trade-off — it takes longer to learn than typing a prompt into a chat box, but it hands the user meaningfully more control once they understand the interface.

The platform is available as a web app and through native iOS and Android apps; it does not offer a downloadable desktop client or local/on-premise installation, since generation runs entirely in Leonardo’s cloud infrastructure.

User Sentiment: What Reviewers Say

Leonardo’s review profile is strong but has a notable pattern worth reporting plainly rather than averaging away. On G2, Leonardo carries a high rating from a comparatively small review sample, with users most often praising ease of use, output quality, and the breadth of available models; the most common complaints relate to credit limitations and cost at scale.

There’s a wrinkle worth flagging directly: independent analysis of a 100-review Trustpilot sample found that a large majority of positive reviews explicitly name a specific support agent, a pattern that strongly suggests many reviews are solicited immediately after a support interaction rather than submitted as unprompted product assessments.

That doesn’t make the underlying sentiment false — Leonardo’s support team is repeatedly described as fast and effective — but it does mean the aggregate star rating may reflect support quality more than it reflects a random cross-section of the paying-customer experience.

Strengths and Limitations

Across research and available third-party testing, Leonardo AI’s strongest demonstrated capabilities are its post-generation editing tools, its custom model training for character and style consistency, and its aggregation of multiple leading image and video models under one subscription.

Its most consistently cited limitations are the unpredictability of token consumption for heavier features, video generation quality that trails dedicated video-first platforms, and a pricing/plan-change structure (immediate upgrades, cycle-end downgrades) that occasionally surprises users during cancellation.

Leonardo AI vs. Midjourney, Adobe Firefly, and Other Alternatives

Compared with Midjourney, Leonardo trades some default aesthetic polish for a web-based interface (no Discord required), built-in editing tools, and clearer commercial IP terms that don’t hinge on a revenue threshold.

Compared with Adobe Firefly, Leonardo is generally cheaper as a standalone subscription, though Firefly has an edge for teams already embedded in Adobe’s Creative Cloud ecosystem and enterprise content-credentialing needs.

Against newer specialists — Recraft for vector and logo work, Flux 2 Pro for photorealism, or Runway and Kling for primary video production — Leonardo tends to be a strong generalist rather than the outright leader in any single narrow category.

Its case is strongest when a team needs one platform that credibly covers image generation, iterative editing, custom model training, and lightweight video in a single subscription, rather than stitching several specialist tools together.

Who Should Use Leonardo AI?

Leonardo AI is well suited to game developers and character designers who need consistent, reusable visual assets across many generations; marketing and content teams that want brand-consistent imagery without hiring a design contractor for every variation; and developers who want a visual prototyping surface that exports directly to production API code.

It’s less suited to users whose primary need is a single, best-in-class photorealistic image with minimal setup — Flux 2 Pro or Midjourney are likely to satisfy that need with less configuration — and it’s not a substitute for a dedicated video generation platform if video is the core deliverable rather than a supporting asset.