Leonardo AI

What Leonardo.Ai is

Leonardo.Ai is a visual creation platform combining image generation, editing, enlargement and video. It brings models developed by its own team together with models from other providers in a shared environment. Its value lies in connecting exploration, selection and transformation, particularly when producing several related assets.

The platform should be distinguished from the selected model. Two images made in Leonardo can come from different technologies and offer different controls. A generated asset also differs from a finished design: an image might supply the raw material for a poster, presentation or video while still requiring wording, structure and review. That distinction helps decide which work belongs in Leonardo and which should continue in a design or editing application. It also makes comparisons more meaningful than treating every result as the output of one uniform system.

Reference documentation: Leonardo.Ai overview.

Origins, investment and Canva acquisition

Leonardo emerged in Sydney in late 2022, initially focusing on assets for games. Investor Blackbird identifies JJ Fiasson as a founder and dates its investment to 2023. The product subsequently expanded into illustration, advertising, concept development and other creative work.

In June 2024 it introduced Phoenix, its first proprietary foundation model. On 30 July 2024 it announced that it was joining Canva. The acquisition announcement envisaged Leonardo remaining an independent platform while expanding research and development. This relationship explains the connection between the tools, but does not merge all their capabilities, accounts and plans into one product. For an existing Canva user, the practical questions concern which assets can transfer and which permissions or subscriptions each stage requires. Corporate ownership and the actual user workflow are related but separate matters.

Reference documentation: Leonardo joins Canva, 30 July 2024 · Canva acquisition announcement · Blackbird portfolio: Leonardo AI.

Proprietary and third-party models

Phoenix marked Leonardo’s move into its own foundation models. Lucid Origin, released on 5 August 2025, was introduced as a versatile model with Full HD output and varied styles. Omni documentation also describes families such as FLUX, GPT and Nano Banana for contextual generation and editing. The available selection changes over time.

Choosing a model means considering the brief: exploring atmosphere, depicting a recognisable product and composing an image containing text involve different criteria. Compare models using the same reference material and a short list of essential conditions. Record which model produced the selected image and which settings were used. A successful result from one model does not establish that another will preserve the same properties, even when both are offered inside the same application and accept similar-looking instructions.

Reference documentation: Phoenix model · Lucid Origin release · Omni Models and Inline Editor.

References and contextual editing

The Omni workflow accepts references and instructions describing how they should combine; current guidance allows up to six images depending on the model. The inline editor supports changes in natural language. Earlier Image Guidance controls, such as style, content or pose references, remain documented as workflows with specific compatibility. Their limits should not be confused with those of newer models.

A useful instruction identifies what should change and what should remain. When adapting a character to another scene, for example, separate identity, clothing, action and background, then inspect each group of elements. If an edit unintentionally changes a tool or sign, returning to an earlier version may be more effective than stacking corrections. Retain authorised references and the selected asset so a later update does not depend on reconstructing decisions from memory.

Reference documentation: Leonardo.Ai — Image Guidance · Omni Models and Inline Editor · Image editor.

Realtime Canvas and Flow State

Realtime Canvas transforms sketches into images while drawing and adjusts how freely the input is interpreted. Flow State offers another form of exploration: it generates a stream of images and narrows a direction through similar variations. In Flow State, consumption applies to generated images, rather than only those saved or downloaded.

Sketching is useful when spatial arrangement matters more than finding the right words. A stream of alternatives helps when the direction is still undecided. Before starting, decide what will end the exploration: three comparable options, a chosen palette or a composition with sufficient room for wording. Without that criterion, more alternatives can increase selection work. Save relevant options early, since changing the description or settings starts another series. This gives exploration a tangible output that can be reviewed with the rest of the project.

Reference documentation: Realtime Canvas · Flow State.

Elements and training on your images

Elements is Leonardo’s implementation of LoRA adaptations: adjustments trained on a collection of images to guide a style, subject or product. Documentation describes preparing the dataset, training and applying the element in later generations. Compatibility depends on the base model; an element is not necessarily transferable to every available model.

Training may make sense when a visual identity will recur across many briefs. A reference may suffice for a one-off asset. The collection should represent what needs preserving, without confusing it with an accidentally repeated background or lighting setup. Reserve different examples for evaluation: another viewpoint, a new scene or a different aspect ratio. If the system only reproduces training compositions effectively, it has not yet demonstrated the variation the project requires. Evaluate the intended range of work rather than simply the most successful initial example.

Reference documentation: Elements training · Elements API training guide.

Video, audio and scene continuity

Leonardo generates video from text or images. Start and end frames, sound, duration and movement controls depend on the model. Motion 2.0 has its own controls and elements, while other models offer different capabilities. The guide specifies MP4 download and warns that some operations described as editing regenerate the video.

For a sequence, validate reference frames first and movement second. A correct opening scene can develop an inconsistent action or change an object during the clip. Review generated sound as well: a phrase, alarm or unexpected effect can change the meaning. Where exact continuity is required, compare correction effort with recording or animating controlled material. Acceptance criteria should cover the beginning, development and conclusion of the action, rather than an attractive thumbnail alone. Viewing the complete clip is a separate review from approving the still image.

Reference documentation: Video generation guide.

3D generation and output scope

The August 2026 guide documents 3D generation through Hyper3D’s Rodin V2. Reference images produce a textured mesh that can be inspected in a viewer and downloaded as GLB. Multiple views can guide the shape.

This workflow supports exploring a visual object from different angles or preparing an asset for another application. A generated mesh is not equivalent to an engineering model with verified dimensions, materials and tolerances. Before reuse, inspect hidden parts, surface continuity and behaviour in the destination program. A convincing front view may leave the back or joints unresolved. Review should follow the intended use, rather than only the first view’s appearance. For a decorative scene, visual consistency may be sufficient; a dimensional explanation needs a source whose geometry can actually be checked.

Reference documentation: 3D generation.

Enlargement, delivery and Canva integration

Pro Upscaler distinguishes Precise and Creative modes with different degrees of detail reconstruction. The official page describes JPG, PNG and WebP inputs and enlargement up to 105 megapixels. The resulting file size does not establish that every detail existed in the original; compare both at a suitable scale before printing.

Edit in Canva transfers an image into a Canva account for further design work. According to the reviewed help page, it is available to individual users, including free accounts, but not Teams, and transfers one image at a time. Check which Canva account is connected. For delivery, retain both the approved image and the design containing wording and corporate assets. A headline can then change without regenerating the illustration or altering details already reviewed. This separates the approved visual from the surrounding communication.

Reference documentation: Pro Upscaler · Edit in Canva.

Blueprints, agent and API integration

Blueprints provides prepared visual workflows combining models, references and settings. The public interface labels Leo Agent as Beta. The API supports generation inside other systems, and the offering describes exporting visual configurations into code. These routes automate different steps and do not necessarily have the same limits as a manual session.

Before integrating, define what triggers work, which files enter, where outputs are stored and who approves their use. For campaign images, automation might prepare variants; editorial approval should be able to stop a defective asset without repeating the entire batch. Keep configuration and consumption records. If a prepared workflow changes its underlying model, review a representative sample before reusing it across the whole series. This gives a repeated process a clear acceptance point rather than equating successful generation with a finished deliverable.

Reference documentation: Blueprints workflows · Public application interface · Developer API.

Plans, devices and teamwork

The offering reviewed on 8 November 2026 distinguishes Free, Essential, Premium and Ultimate, alongside team and API plans. Access is advertised for the web and mobile devices. Plans determine models, queues and private generations; unlimited relaxed generation applies to selected models.

Teams includes shared assets and tokens, collections and common models. To choose a plan, estimate an entire series: exploration, corrections, enlargement and video where needed. A team also needs to decide who retains the original and who authorises a new version. Sharing assets supports continuity, but without clear names and criteria it can leave several apparently final images. Organise work by project and review status rather than generation date alone. This makes handoff and future revisions easier to assess alongside the subscription’s technical limits.

Reference documentation: Individual plans · Leonardo for Teams.

Privacy, rights and commercial use

The help centre permits commercial use subject to terms and distinguishes public from private generations. Private options require a paid account. The terms allocate rights differently between free and paid subscribers and restrict provider use of private content to contractual purposes, without additional training unless expressly authorised.

Before uploading material, check the account terms and each asset’s visibility. A licence to use an output is not exclusivity and does not settle permissions for faces, brands or input images. To train a corporate style, use a collection whose origin and authorisation can be documented. The ability to create a variation of a product or person does not establish that the variation is authorised for advertising. Decide based on the particular asset and intended purpose, including the source references as well as the final output.

Reference documentation: Leonardo.Ai — Private and public images · Terms of Service · Commercial usage.

Practical applications in OSH

It can support illustrations for safety communication and conceptual scenes for learning materials. Its relationship with occupational safety and health depends on the supplied content and technical review: the platform produces visuals but does not establish that an operation is safe or that a depiction matches the actual workplace.

It can support an illustrated sequence, variants of a training scenario or assets with a shared appearance across modules. In OSH, distinguish decorative background from elements teaching a behaviour. An obstacle’s size, a load’s position or a protective measure is not merely an aesthetic detail.

Present scenarios as examples rather than workplace photographs. When a real location must be represented, use authorised assets whose correspondence with that location can be checked. Keep an inventory of the references so later updates can follow the same visual and editorial decisions.

Worked example and result checks

Example: develop a consistent series of illustrations explaining why walkways must remain clear. The team defines the scenarios, reviews each depicted obstruction and adds the specific actions agreed for the site. Workers can then discuss how the message applies to their own routes.

1. Define three scenes and their learning points: recognising an obstruction, reporting it and keeping the route clear under workplace instructions.
2. Prepare authorised references and a list of elements that must remain constant, such as route, lighting and approved signs.
3. Generate and compare scenes together. Look for changes that confuse the sequence, not only defects within individual images.
4. Check that the result does not normalise inappropriate practice. Replace a scene requiring extensive corrections with a clear diagram.
5. Add approved instructions as editable text and ask someone to explain the sequence in their own words.

This proposed exercise supports evaluation of the material; it does not claim that the images have been tested with actual workers.

Limitations and conditions of use

Enhancement or editing may introduce details absent from the original. Review entire images, especially machinery, protective equipment and signs. Keep a reference version and do not treat an altered scene as evidence of actual workplace conditions.

Check visibility before uploading references. Documentation distinguishes public and private generations, with private options linked to particular accounts. Do not use personal images or workplace information assuming that every creation stays confidential.

A reference neither transfers image rights nor guarantees exact preservation of identity. It can also introduce unwanted characteristics. Check that people, equipment and settings are represented appropriately, without relying on stereotypes to explain a risk.

Reference documentation: Leonardo.Ai — Image Guidance · Leonardo.Ai — Private and public images.

When to choose it and how to complement it

Leonardo AI fits development of a series using references and editing options. Midjourney offers another visual exploration workflow; choose using a representative project scene and the correction effort required. Canva can subsequently organise the sequence with wording and corporate assets.

Include a character or background change in the trial. Check what changes unexpectedly and whether the team can maintain consistency during future updates. A good first image is insufficient if the remaining sequence cannot be maintained.

Frequently asked questions

Does an image reference guarantee an exact reproduction?

No. It guides generation, but the result must be compared against the elements that need preserving.

How should an illustration series be reviewed?

View it as a whole: check continuity of characters, equipment, spaces and actions, alongside each individual scene.

What should be checked before uploading a photograph?

Usage rights, authorisation concerning people or locations, and the project’s actual visibility settings.