Leonardo

The free day here is a real one, so the decision comes down to how many images you need a day and on whose models.

Who it is for. For those who want to browse other companies’ top image models in one place without subscribing to each one: Google’s Nano Banana, OpenAI’s GPT Image, and Kling for video are listed alongside its own Lucid and Phoenix. For designers and game creators who need their own style: Leonardo trains a model on 10–50 of your images.

Who else. For those who want to try it without a card and are not in a hurry: 150 tokens arrive every day, which is around a dozen images on its own models, adding up to more than three hundred per month.

Who it is not for. People who want privacy for free: on the free plan, all images are public, owned by the vendor, and cannot be deleted.

Suitable: section 3 takes you from sign-in to your first image. Midjourney

What vibecoding isStarting from scratch: how to set tasks for agents with no coding experience

Three doors: its own models, other companies’ top models, and video. All three use the same currency, tokens, and the price appears on the button before every generation.

Models. It has four of its own: Lucid Origin and Lucid Realism for images, Phoenix in two versions, and Motion 2.0 for video. Its API documentation listed around forty external models from a dozen companies as of 09/07/2026, including Google’s Nano Banana, OpenAI’s GPT Image, and Kling; the full list is in the FAQ. Only external models appear in the recommended block on the app’s first screen, and the vendor itself describes Leonardo as a place where “the world’s best models are brought together in one system.”

Editing. Edit an image with words in the editor: describe the change, and the model redraws what is needed. The old canvas with masks and a brush remains, but the vendor itself calls it the previous generation. Pro Upscaler raises an image to 105 megapixels, the background is removed with a button, and a 3D model can be made from an image using external Rodin V2 from 400 tokens.

Video. There is one in-house video model, Motion 2.0, which animates a finished image by default. External Kling 3.0, Veo 3.1, and Seedance make videos with sound from 3 to 15 seconds long; its own Motion 2.0 goes up to 720p, Veo 3.1 up to 4K. Only paid users can upload their own image for video, while free videos have a watermark.

Your own style. The model learns from your images: from 10 to 50 files, preferably square at 1024 pixels, creating an add-on over Flux Dev. Essential includes ten such models per month. This is Leonardo’s main move for games and brands: set a style once, then create in it.

What the vendor itself shows. The loop above the card comes from a video on the character generator page, published 05/12/2026: a home screen with a prompt bar and several modes, a shelf of ready-made Blueprints scenarios, image selection, and a scene wizard with several characters; at the end, a “Generate 240” button priced in tokens and an “Edit in Canva” menu. We did not run Leonardo ourselves for this task.

Nothing to install: the website and two apps from the vendor.

Step 1. Open app.leonardo.ai and sign in with Canva, Google, Apple, Microsoft or email, that is how the sign-in screen looks on 07.09.2026.

Step 2. The first 150 tokens arrive immediately and refresh every 24 hours. Write an English prompt in the bar at the top, choose a model, and the token price is on the button. Everything you create for free is visible to everyone in the feed, quality is basic only, and you cannot delete an image without a subscription.

Step 3. A subscription from $12 a month is set up through Stripe or PayPal.  

Apps. On the App Store the app is published by the vendor itself: 4.87 from 1,186 ratings, updated on 26.08.2026. On Google Play it has more than 10 million installs.

Steps verified on Sep 7, 2026 against the sign-in screen in a live browser, the vendor's help center and the store listings

In image arenas, Leonardo sits in the second half of the list, and its best model there is a year old.

Arenas. In Artificial Analysis as of Sep 7, 2026, Leonardo's best model, Lucid Origin Ultra, ranks 79th out of 157 with a score of 923; leader GPT Image 2 from OpenAI has 1,178. In LMArena, the same Lucid Origin ranks 68th out of 76 with a score of 1,013 from 285 thousand votes. There are no Leonardo models newer than August 2025 in the arenas, while the leaders of those arenas are available inside Leonardo itself as third-party models.

What this means. The vendor does not disclose absolute numbers for 2026, only percentage growth. Of a16z's list of the top 100 consumer AI products from Mar 9, 2026, only three standalone image generators remain: Leonardo, Ideogram, and CivitAI. The list's authors explain this by saying they serve specific creative communities rather than compete in general generation. Leonardo's place among image tools is shown below the card title, while the full tool ranking is on the general page.

The free 150 tokens per day are real, but a subscription from $12 buys tokens, not images, and they expire.

Tokens. There is no single rate: the price depends on the model, size, and quality, and is shown on the button before generation. The vendor's only published table prices an image at 1 token, but it covers 2024 models; in the vendor's own footage, an image on GPT Image 2 costs 6.97 tokens, while an upscale to 33 megapixels costs 573. 2026 hands-on tests put its own models at 8-10 tokens per image, so a free day gives about a dozen images, not 150.

Three plans. Essential for $12 gives 8,500 tokens per month, or 850 to 1,000 images; when they run out, the account drops to the free 150 per day. Premium for $30 gives 25,000 and, after zero, enables an unmetered slow queue, but only on Leonardo's own models. Ultimate for $60 gives 60,000 and the same queue for video too. Third-party models are almost entirely excluded from the unmetered queue: the help lists 16 names and the pricing page lists 17, while a footnote on that page allows three third-party models into the queue: Hailuo 2.3, Hailuo 2.3 Fast, and Wan 2.6. The annual price has not changed.

Annual billing and token bank. Paying annually lowers the monthly price to $10, $24, and $48. For paid users, unused tokens accumulate in a bank up to three monthly allowances; free users have no bank, and canceling a subscription burns the entire bank, which reviews complain about. Token packs can be purchased only with an active subscription; in the App Store, a 5,500-token pack costs $13, making top-ups one and a half to three times more expensive per token than subscription tokens.

Mobile markup. Through the iPhone app the same plans cost $15.99, $39.99 and $79.99, a third more than on the site. Subscribing on the site is cheaper.

The vendor accepts cards through Stripe and PayPal. The service terms prohibit resale and simultaneous login under one account.  .

Through Canva. The Essential plan is marked "included with Canva Business" on the pricing page: anyone who already has this Canva subscription gets Leonardo's entry-level plan with it.

The main question here is not lawsuits, of which there are none, but the free plan: images made on it belong to the vendor.

Whose images. On paid plans, they are yours: the terms transfer all rights to you at the moment of creation. On the free plan, rights arise with Leonardo, while you receive a perpetual license, including for commercial use. Free images are public, other users can remix them, and you cannot delete them without a subscription.

Data and training. Data is stored in the US, Australia, and any other country at the vendor’s choice; the data operator is Sydney-based Leonardo Interactive, identified in the privacy policy as a Canva brand. The company has never directly said what Phoenix and Lucid Origin were trained on: in 2023, it spoke of synthetic data and Creative Commons; in 2024, of “licensed, synthetic and publicly available data”; in 2025, of careful dataset selection. No copyright lawsuits against Leonardo were found as of 07.09.2026.

Filters. In April 2024, a 404 Media journalist used Reddit prompt suggestions to generate intimate images of celebrities without their consent through Leonardo; the vendor closed the loophole within two weeks and published a safety statement on 06.05.2024. Since then, filters for free users have been stricter than for paid users, according to the vendor’s help article.

Who owns it. Canva since Jul 30, 2024: the deal amount was not disclosed, while the press estimated it at more than 300 million without naming the currency. The company then promised “the same plans and independent platform,” and prices have remained in place since at least August 2025. In March 2026, Canva accelerated integration and distributed the 150-person team across its AI departments, retaining a small group for Leonardo, while denying layoffs; the apps have been updated since then, most recently on 26.08.2026. The service has no public status page, and it does not respond.

The fork is not about who draws more beautifully, but what you are buying: one model with its own taste or a storefront of other models with a free tier.

Different checkout. Leonardo gives 150 tokens free every day, with subscriptions from $12. Midjourney has offered no free trial since Mar 28, 2023, starts at $10 a month, and sells GPU minutes rather than tokens.

Different models. Midjourney has one proprietary model, V8.2, and prohibits any third-party API. Leonardo has four proprietary models and nearly forty external ones, including arena leaders. Yet Leonardo’s proprietary model ranks 79th out of 157 in the arenas, while Midjourney does not submit its new model there at all, so their quality cannot be compared numerically.

Who else is nearby. Among its international catalog neighbors, Recraft, Ideogram and Krea.

Over four years, the Sydney game-asset generator gained its own model, became a Canva brand, and turned into a storefront for other models.