Leonardo.Ai is a browser-based generative AI platform that turns text prompts, reference images, and custom-trained models into production-ready images and video. It combines in-house Phoenix, Lucid, and Motion models with licensed third-party engines, and serves creators, game studios, marketing teams, and developers through both a web app and a visual-first API.
Used this tool? Rate it
Used this tool? Rate it
Leonardo.Ai presents itself as "The Creator-First Generative AI Platform", a cloud service spanning images, art, and video. In practice it is a browser-based studio where you type a prompt or upload a reference, pick from a shelf of generative models, and get back an image or a video clip — then keep editing that result inside the same workspace rather than exporting it to another tool.
What separates it from the single-model generators that dominated the first wave of AI art is the deliberate refusal to bet on one engine. The company states plainly that "our platform isn't restricted to a single generalized model; instead, we have an assortment of both general and finely-tuned models", each built to excel at a different kind of output. On top of that sits the feature that has defined the product since its early days: you can train your own model. Users can upload a small collection of images (typically 10-20) to train a unique model that recreates a specific style, theme, or subject, picking up nuances like colour patterns, composition, and aesthetic concepts.
The operating entity is Leonardo Interactive Pty Ltd T/A Leonardo.Ai (ACN 662 209 485), an Australian company, and its Terms of Service were last updated on 19 January 2026. The company was founded in Sydney in late 2022, initially aimed at game developers who needed concept art and asset variations quickly — an origin that still shows in the product's unusually deep control surface.
This is the part of Leonardo's story most third-party listings get wrong, because it has moved twice.
Canva announced its acquisition of Leonardo.Ai on 29 July 2024. The framing at the time was explicit: "Leonardo will continue to operate independently and focus on rapid innovation, model research, and product development." Reporting placed the deal above USD 300 million. For roughly eighteen months that promise held — Leonardo kept its own brand, its own pricing page, and its own roadmap, while contributing model technology into Canva's Magic Studio.
The ownership is no longer implicit. The privacy policy opens by identifying the company as "a Canva brand", which is the clearest first-party confirmation of the ownership relationship.
Then, in March 2026, the arrangement changed. Canva confirmed it was accelerating the integration of Leonardo into its wider AI roadmap. A Canva spokesperson stated that "A small team will continue to work on Leonardo as a standalone product, and everyone else will move into new teams across Canva". Australian outlets reported that staff at Leonardo's North Sydney office had been briefed on a restructure; Canva rejected accompanying claims of imminent job losses, with its head of corporate communications stating that no job losses were discussed or communicated to the team. Those competing accounts have not been fully reconciled in public.
What this means for a prospective user is narrower than the headlines suggest. As of this writing the platform is live, sells subscriptions, publishes its own pricing, and ships new models. But the team behind it is materially smaller than during the standalone era, and long-horizon planning should account for the possibility that capabilities migrate into Canva's own surface over time. Anyone building a business-critical pipeline on it should read that alongside the platform's own reserved right to change the service, discussed later in the limitations section.
Leonardo runs two model families of its own — Phoenix and Lucid on the image side, Motion on the video side — and licenses a rotating set of frontier third-party engines alongside them. The commercial line between the two is visible in the pricing terms: the unlimited relaxed pool currently includes Lucid Origin, Lucid Realism, Phoenix 1.0, Phoenix 0.9, Motion 1.0, Motion 2.0, and Motion 2.0 Fast. Third-party engines — names like Veo, Kling, Sora, Seedream, Ideogram, Nano Banana Pro, and FLUX variants appear across the platform — are billed per generation and excluded from that unlimited pool.
That distinction is not a detail. It is the single most useful thing to understand before choosing a plan, because it determines whether your workflow has a cost ceiling or a running meter.
Training your own model is Leonardo's most durable differentiator. Feed it ten to twenty images of a character, a product, an art direction, or a texture family, and it produces a fine-tuned model you can generate from repeatedly. For a studio that needs two hundred assets to look like they came from the same artist, this is the difference between a toy and a pipeline. Model slots are allocated by tier, from ten on the entry paid plan up to fifty at the top.
The platform's editing tools are built around not losing what already works. Leonardo describes the goal as editing images and video while preserving the elements that matter — characters, composition, style, and intent — so you refine rather than reroll. In practice this covers inpainting and outpainting on a canvas, image-to-image transformation, character reference to keep a subject consistent across a set, and structural controls that lock pose and layout.
Video is a first-class product line rather than a bolted-on extra, with its own Motion model family, its own queue limits, and its own unlimited tier at the top plan. It supports image-to-video workflows in addition to text-driven generation. It is also, by consistent third-party assessment, the less mature half of the platform: reviewers rate image generation as best-in-class while describing video output and sequential consistency as still catching up.
Resolution and clarity enhancement is treated as a delivery step — preparing assets for print, high-resolution displays, and professional handoff. Upscaling is a token-billed operation like any other generation, which matters when budgeting a large batch.
In February 2026, alongside a brand refresh, Leonardo launched what it calls a Creative Engine API. Its pitch inverts the usual developer workflow: "Design media generation in a visual interface, see the output in real time, then export the exact API code to production." You dial in the result with sliders in the web app, then export runnable code with every parameter already set, avoiding the translation step where a designer's intent gets lost in a JSON payload. The API supports workflows like text-to-image, image-to-image, and image-to-video generation, all in one place, plus LoRA training and custom presets.
Game art and game-ready assets. This is where the platform's depth pays off most directly, and it matches its origin story. One hands-on reviewer rated it the best image generator tested for game art and game-ready assets, crediting not the raw beauty of the pictures but the production controls surrounding them — pose locking, character reference, inpainting, custom model training, and PBR texture map generation that imports into a 3D engine.
Marketing and campaign visuals at scale. The company's published case studies point here: Gamepack reported halving production spend while maintaining quality, Ducati used the platform to power a global campaign producing personalised shareable visuals for riders, and Springbok used the video models to shorten time-to-market for motion assets.
Brand-consistent content production. Train a model on your brand's existing visual library, then generate against it. This is the workflow that makes custom training worth its price for agencies and in-house teams who need output that survives a brand review.
Concept exploration and pre-visualisation. Fast iteration across styles and compositions before committing budget to photography, illustration, or 3D production. The free daily token allocation makes this viable to test before any spend.
Developer-embedded generation. Through the API, generation becomes a feature inside someone else's product — dynamic game worlds, personalised commerce imagery, or user-facing creative tools. The company reported 50 per cent year-on-year growth in its user base and a 200 per cent increase in ARR with business customers, figures that are vendor-reported and not independently audited.
Leonardo runs a freemium subscription with a token currency, plus a completely separate API billing track.
| Plan | Price (ex. tax) | Fast Tokens | Rollover bank | Concurrency |
|---|---|---|---|---|
| Free | $0 / month | 150 per day | — | — |
| Essential | $12 / month | 8,500 per month | 25,500 | 2 |
| Premium | $30 / month | 25,000 per month | 75,000 | 3 |
| Ultimate | $60 / month | 60,000 per month | 180,000 | 6 |
Annual billing is advertised at up to 20% off. Generation queue limits scale with the tier, from 5 on the entry plan to 20 at the top.
Tokens are the platform's universal unit of account, deducted for image generation, video generation, post-processing such as background removal, and upscaling. Crucially, a token does not map to a fixed output. The number of tokens used is based on the computational intensity of the action you are taking - the more GPU load required, the more tokens are charged. Higher resolution, heavier models, and enhancement passes all multiply the cost of a single button press, which is why newcomers routinely burn a day's free allocation faster than they expected.
Paid plans add a rollover bank so unused tokens carry into the next cycle up to a per-tier ceiling. Separately, top-up packs can be purchased; these do not expire, are consumed only after rollover and fast tokens, and require an active subscription to be usable — meaning they are not a way to keep access after cancelling.
Premium and Ultimate plans do not hard-stop when the token pool empties. Relaxed Generation lets you keep generating images and videos without extra token cost, but at a lower priority than the main queue. Wait times fluctuate with overall platform demand and concurrency is reduced. Premium gets relaxed image generation; Ultimate adds relaxed video. Only the eligible first-party models listed earlier qualify.
The API is billed independently of subscriptions. New API accounts receive $5 in free credit — "Every new API account starts with $5 in free API credit, allowing you to test the API, explore features, and validate your integration before spending anything." From there it is pay-as-you-go with non-expiring credit and automatic top-ups, or a custom contract for higher concurrency and volume discounts.
This section deserves close reading, because Leonardo's terms are tiered in a way that catches free users out.
Paid subscribers retain full ownership, copyright, and other intellectual property rights of images that they generate, while on the free tier Leonardo.Ai holds rights to use, reproduce, modify, and distribute any images you create. Free users are granted a non-exclusive, royalty-free licence for commercial use of their own creations, and their public assets can be accessed and used by other users through platform features. The practical reading: if the output has commercial value to you, the $12 tier is not optional.
For API-generated content the position is cleaner still — Leonardo assigns to the user all rights, titles, and interests in the outputs generated from their inputs, and confirms API-generated images may be used commercially.
Yes, by default on the web platform — and there is an explicit opt-out. The privacy policy states that user content and related account data may be analysed to improve the service and to train the company's algorithms, models, and AI products using machine learning. The escape hatch is tied to payment: if you do not want your Content to be used for machine learning, you can generate Content privately with any paid subscription plan, and that private content will not be used by Leonardo.Ai for model training.
On the API side the commitment is stricter still: input images are never used for training, and images generated through the Leonardo API are not used for training either.
Separately, the company discloses that it collects information that is publicly available on the internet to train its models and conduct research, acknowledging that personal information may be incidentally included in such datasets. That places Leonardo in the same disclosed-web-scraping category as most foundation model developers, and is worth knowing if provenance matters to your organisation.
A strong fit for game developers and studios needing consistent asset volume; freelance designers and marketers producing campaign visuals; creators who want more control than a prompt box offers; teams needing brand-consistent output via custom models; and developers embedding generation into a product.
A poor fit for anyone whose priority is the highest artistic ceiling on a single photorealistic image, where reviewers still favour rivals; users who want simple per-image pricing without token arithmetic; free-tier users who need copyright ownership; workflows that require permissive content handling, given the platform's strict moderation; and anyone requiring long-term vendor stability guarantees given the ongoing integration into Canva.
The primary surface is the browser-based web application, with iOS and Android apps also available, and the API as a third route. There is no requirement to use Discord, which is a meaningful workflow difference from some competitors. Access is account-gated, and the platform emphasises that no coding skill is required to begin.
You pay for failures. Independent reviewers single out one recurring grievance above all others: failed and moderation-blocked generations still consume tokens. Because the platform's own safety filter can reject a prompt after the tokens are committed, the true cost per usable image is always higher than the nominal token price. Budget for waste.
Moderation is aggressive. The most common substantive complaint in user reviews concerns the content safety filter blocking benign professional prompts. One reviewer described the experience directly: prompts that are clearly non-sexual, cinematic, and professional fail with vague "content safety" errors, with no indication of what triggered the block. For most professional users the guardrails are acceptable or even desirable; for anyone working in genre fiction, horror, or fashion imagery they are a real friction point.
The learning curve is real. Aggregated review data describes usability as mixed, driven by a steep learning curve for beginners despite praise from experienced users. The density of settings that makes the platform powerful also makes the first session confusing.
Cost is hard to forecast. Because token consumption scales with computational load rather than output count, predicting a monthly spend is genuinely difficult — a complaint that recurs across review platforms.
Video trails image. Reviewers consistently rate the image side as best-in-class and the video side as still maturing, particularly for sequential and character-consistent output.
The vendor reserves broad discretion. The terms state the company may add, change or remove features or functionality to its services; modify or introduce limitations to storage or other features; or discontinue the services altogether at any time — with a commitment to migrate to a substantially similar service or issue a pro-rata refund if a paid service is discontinued mid-subscription. Combined with the Canva integration, this is the clause enterprise buyers should weigh.
Accounts expire. The minimum age is 13, and accounts inactive for 12 months or more may be closed and permanently deleted, taking generated content with them. Export anything you care about.
On Trustpilot the platform carries a 4.6 score across 1,176 reviews, and the company has replied to 74% of negative reviews. Aggregated data from other review platforms points the same direction: Leonardo.Ai earns a broadly positive profile across image generation quality, feature depth, and cost-effectiveness, drawing on G2 and Capterra reviews spanning September 2023 to July 2026.
One caveat deserves airing, because at least one analyst has raised it directly: a very large share of the positive Trustpilot reviews specifically name a support agent, which suggests reviews are solicited after support interactions. That analysis concluded the 4.6 rating may tell you more about the help desk than the product itself. The reviews appear genuine and detailed; the point is about composition, not authenticity. Read the score as strong evidence of responsive support and weaker evidence of average product satisfaction.
Negative themes cluster tightly and consistently: the token-on-failure billing, the moderation filter, subscription renewal clarity, and — in one documented enterprise case — a serious billing and cancellation failure. On the positive side, image quality, model variety, custom training, canvas editing, and the generosity of the free tier draw repeated praise.
Midjourney — generally held to have a higher artistic ceiling, particularly for photorealistic and editorial imagery, but with no meaningful free tier and fewer editing and training tools.
Canva — now Leonardo's own parent, and the more integrated choice if your work ends in a layout, a deck, or a social post rather than a raw asset.
Adobe Firefly — the choice when commercially-cleared training data and Creative Cloud integration outweigh raw flexibility.
Stable Diffusion (self-hosted) — maximum control, no content filter, no per-generation cost, at the price of running your own infrastructure and assembling your own toolchain.
DALL·E via ChatGPT — simplest conversational path to a decent image, with far less control over the result.
The honest framing: Leonardo competes on control and breadth rather than on any single dimension. If you need one beautiful picture, look elsewhere. If you need two hundred consistent assets, a repeatable style, and an API to ship it through, that is the problem it was built for.
Yes. The free tier provides 150 Fast Tokens per day with no credit card required, and the company states the free offering has no expiry date. The limits that matter are not just quantitative: free creations are public rather than private, quality settings are basic, custom model training is not included, and — most importantly — you do not receive copyright ownership of your outputs.
Yes, but the terms differ sharply by tier. Paid subscribers retain full ownership and copyright of what they generate. Free-tier users receive a non-exclusive, royalty-free licence for commercial use, but Leonardo.Ai retains rights to use, reproduce, modify and distribute those images, and other users may access public assets. Content generated through the API is assigned to the user and may be used commercially. If commercial certainty matters, use a paid plan.
By default on the web platform, user content may be analysed and used to train the company's models. There is an explicit opt-out, but it requires payment: generating privately on any paid subscription plan keeps that content out of model training. The API is stricter by default — neither input images nor API-generated images are used for training.
No, not in the ownership sense — the privacy policy identifies it as a Canva brand, following an acquisition announced on 29 July 2024. Operationally, it ran as a standalone business for about eighteen months, but in March 2026 Canva confirmed it was accelerating integration, with a small team continuing on Leonardo as a standalone product and other staff moving into teams across Canva. The platform remains live and sells subscriptions, but it is no longer a fully independent operation.
Tokens are a general-purpose currency deducted for generation, post-processing and upscaling. A token does not correspond to a fixed image size or action — consumption is based on the computational intensity of the task, so higher resolution, heavier models and enhancement passes each multiply the cost. This is the main reason monthly spend is difficult to forecast, and the most common cost complaint in user reviews.
On Premium and Ultimate plans, once the token pool is exhausted you can keep generating at no additional token cost via Relaxed Generation, at lower queue priority and with reduced concurrency. Wait times vary with platform demand. Premium covers relaxed image generation; Ultimate adds video. Only eligible first-party models qualify — licensed third-party engines are always billed per generation.
Yes, and this is the platform's most consistently criticised behaviour. Independent reviews report that failed and moderation-blocked generations still consume tokens, meaning a prompt rejected by Leonardo's own safety filter costs you the same as a successful one. Plan on a real cost per usable image that exceeds the nominal token maths.
No coding is required for the web app — the platform states it is designed for creators of all backgrounds and skill levels. Separately, a full API is available with its own billing: new accounts start with $5 in free credit, and the visual-first workflow lets you design generation settings in the interface and export runnable production code supporting text-to-image, image-to-image, and image-to-video workflows, plus LoRA training.
Four stand out: tokens are consumed by failed generations; the content moderation filter blocks benign prompts more often than most users expect; video generation and character consistency lag behind still-image quality; and the terms permit the company to change or discontinue services at any time, with pro-rata refunds if a paid service ends mid-subscription. Accounts inactive for 12 months or more may also be permanently deleted along with their content.
Creators and teams who need volume and repeatability rather than a single showpiece image: game studios producing consistent asset sets, marketers running campaign variations, designers who want canvas-level control over results, and developers embedding generation into their own products. Users seeking the absolute highest photorealistic quality on one-off images, or who prefer simple per-image pricing, will likely be better served elsewhere.
