A creative workspace for reference-guided images, natural-language edits, ideation, upscaling, collections, and model-specific video.
Leonardo.Ai
Explore features, practical uses and pricing below.
Leonardo.Ai is a creative workspace for generating images, editing them, exploring visual directions, and turning selected images into video. It combines its own models with supported partner models, reference-based generation, natural-language edits, an image library, and enlargement tools. The practical attraction is the ability to develop an asset through several stages: find a direction, make controlled variations, refine a promising result, and prepare it for a design or motion project.
It suits illustrators, game and entertainment designers, marketers, and creators who need several related visuals rather than one isolated image. A writer might develop a character's appearance; a designer might explore an advertising scene; a filmmaker might prepare a reference frame before animation. Leonardo provides tools for those tasks, while the creator remains responsible for the visual decisions, continuity, typography, and final production checks.
The image generator accepts text descriptions and supported image inputs. Model choice affects the available settings and the kind of generation or editing workflow you can use. The platform offers styles and presets, but a preset is a starting direction rather than a complete design brief. A useful prompt describes the subject, scene, viewpoint, lighting, and the intended visual treatment.
For a fictional product campaign, specify the bottle's position, the surrounding material, and where the composition should leave space for a headline. Generate a small selection and judge the overall arrangement before pursuing fine detail. If every result crowds the subject against an edge, change the composition request instead of repeatedly adding more adjectives. Keep a short record of the model and prompt used for the direction you choose.
Separate conceptual imagery from material that must reproduce an actual product accurately. A generated bottle can suggest a lighting style, but the label and shape may need to come from a real photograph in the final advertisement. Leonardo is useful for exploring a setting and mood; the production brief determines which details may change and which must remain exact.
Image Guidance provides reference controls for aspects such as style, content, character, depth, edges, sketch, and pose in compatible workflows. Newer reference-based models also support combining uploaded images through a natural-language instruction. The supported number of images and the available controls depend on the model and workflow, so reference settings should be checked before building a repeatable process.
Different references serve different purposes. A character sheet may establish appearance, a room image may establish the environment, and a sketch may establish the placement of the subject. Explain those roles explicitly. Asking the model to copy everything from all three sources can create an ambiguous brief. For a recurring character, keep identifying features consistent in both the reference and the text description.
Review the details that matter across the resulting set: clothing fasteners, hair shape, proportions, or the placement of a distinctive accessory. Similar overall style is not the same as continuity. If one element drifts, revise that part of the request or the reference selection before generating a new batch. An approved reference set makes a stronger foundation for related scenes than a collection of unrelated attractive images.
Flow State is designed for visual ideation. A prompt produces a stream of images, with options for vibe, lighting, shot type, and color theme. More Like This explores stylistic variations around a selected result. The guide also describes saving useful images and controlling scroll-based generation so the exploration does not continue indefinitely.
This is useful when the brief is still open. For example, explore whether a children's adventure scene should feel like a paper illustration, a softly lit miniature, or a bold graphic composition. Choose a few directions that differ meaningfully rather than saving every pleasant result. Compare them against the audience, the page layout, and the emotional tone of the story.
Once a direction is selected, move from discovery to deliberate refinement. Keep the approved image as a reference and write down what made it suitable: perhaps the low viewpoint and limited palette rather than the incidental objects. A stream of ideas can reveal possibilities, but a production asset needs a stable brief. Stopping the exploration at the right point is part of using the feature effectively.
Describe with AI analyzes an image and produces a descriptive prompt. Leonardo explicitly distinguishes this from image referencing: the selected image is described in words rather than automatically supplied as a visual reference for the next generation. You can edit the resulting description and use it as an idea for another image or video.
The distinction matters when you need a mood rather than a close reproduction. A photograph might suggest muted colors, side lighting, and a sparse interior. Keep those observations and replace the subject and setting for your own project. If preserving a specific character or object is important, use the appropriate reference workflow instead of assuming the description will carry all identifying details.
Treat the generated wording as an editable draft. It may describe a visible surface correctly while guessing the material or overlooking a small object. Remove unnecessary details before using the prompt. A concise description that captures the important visual relationships is often easier to revise than a long list of everything visible in the source.
The Inline Editor supports natural-language changes using compatible Omni models. You can open a generated image, add supported references, and describe the change you want. The documented workflows include adjusting image content, combining reference elements, and editing text. Images created with other models can also be brought into this editing process.
Make revisions around one purpose at a time. If the subject is well placed but the background is too busy, request a simpler background while identifying the subject details to preserve. Review the entire output afterward. An edit aimed at the background may still affect reflections, shadows, or an edge that crosses the subject. Keep the version before the edit so the result can be compared directly.
Text editing needs a close check at the final display size. A heading may look convincing at a glance while containing an incorrect character or inconsistent spacing. For a design requiring exact typography, use the generated image as visual material and finish the editable text in a layout tool. Natural-language editing can help develop the composition, but final production requirements still determine how the asset should be assembled.
Canvas Editor provides inpainting and outpainting around a generation frame, with masking and erasing tools for local changes. Leonardo's guide continues to describe it while recommending the newer editing experience for many prompt-based transformations. The two approaches provide different ways to express an edit: a visible region on a canvas or a written instruction with supported references.
A selected-area workflow can help when the problem is clearly localized. If a scene is suitable except for an unwanted object near the edge, work on that region and include enough surrounding context for the replacement to connect with the scene. For an expanded composition, examine the joins between the original image and the new area. Horizon lines, patterned surfaces, and repeating architecture can reveal a mismatch.
Choose the editing approach that expresses the request most clearly. A written instruction may be convenient for changing an overall setting; a mask may make the location of a small correction unambiguous. Neither approach removes the need to inspect the result. Keep important versions separate so a local improvement does not accidentally replace a stronger overall composition.
Leonardo's Pro Upscaler offers Precise and Creative modes. Precise is intended to retain fidelity to the source, while Creative reconstructs detail with adjustable creativity strength. The tool can work with generated or uploaded images and offers export options, including a supported Canva handoff. Enlargement prepares an image for a larger delivery size, but reconstruction can also change what the image appears to contain.
Choose the mode according to the asset's purpose. A fictional illustration may benefit from reconstructed texture; a product image with an exact surface pattern requires closer fidelity. Compare the enlarged result with the original, especially around faces, small labels, decorative markings, and repeated shapes. More pixels do not automatically make an altered detail acceptable.
Enlarge after approving the composition, rather than spending the entire budget on rejected directions. Check the actual placement in the intended design: a large poster and a small thumbnail have different requirements. Keep a high-quality master and make destination-specific copies afterward. That preserves the option to crop a different version without repeatedly processing the same source.
The video workspace supports text and image generation with model-dependent features. Model labels identify capabilities such as start or end frames and audio. Duration, dimensions, quality, motion controls, and sound support vary. Some partner models are restricted by access level or usage allowances, so the presence of video tools does not mean every model is included in every account.
For an image-led shot, approve the starting frame first and describe the intended motion. A slow camera move across a prepared illustration asks a different question from a character performing several actions. Start with a short, clear movement that supports the story. Check whether the face, clothing, and important props remain usable as the shot develops.
Listen to generated sound when the chosen model produces audio. A clip may include ambience or dialogue that does not match the intended scene. Keep the visual decision and the soundtrack decision separate during review. Save the acceptable shot to the project collection, then assemble it with other footage in the editing environment appropriate for the final video.
Collections organize images and videos into projects and nested groups. You can add generations to a collection during creation or organize existing material from the library. This helps distinguish approved assets from experiments, especially when several directions share similar prompts.
A practical structure is to keep references, selected stills, edited versions, and motion candidates in separate groups. Name the collection around the project rather than the model, since a finished campaign may use more than one model. Save the source that informed an approved result so another person can understand why it belongs in the set.
The Edit in Canva feature sends an image to a connected Canva account for layout work. The documented handoff is available to individual users, requires a Canva account, and sends images individually rather than as a batch. It is useful for adding editable titles, logos, and other design elements after choosing the visual. Team users should check their separate workflow rather than assuming the individual-account handoff applies.
Suppose a bookstore needs three illustrations for a fictional reading adventure. Write a brief defining the main character, palette, setting, and spaces required for event details. Use Flow State to explore a few broad directions, then choose one and establish a reference set. Generate the three scenes with the same identifying character features and review them side by side.
Use the Inline Editor or a selected-area edit to correct a distracting prop. Check the character and important edges again after each revision. Place approved results in a project collection, enlarge only those that need larger delivery, and finish the event title and date as editable text in the layout tool. If one image becomes a motion teaser, specify a restrained action and inspect continuity through the entire clip.
The workflow keeps discovery, continuity, local refinement, and delivery as separate decisions. Each Leonardo tool has a particular purpose in that sequence, which makes it easier to identify the stage that needs revision when a result misses the brief.
The pricing page provides free, paid individual, team, and API access. Paid tiers differ in token allowances, model training, queue limits, privacy, and selected relaxed-generation access. Unlimited relaxed generation is tied to specified models and plan conditions. API usage follows a separate access and billing structure, so a creative subscription should not be assumed to cover an application's API calls.
Generation visibility is also significant. Personal private-generation controls require paid access; team generations have separate internal visibility settings. Choose the appropriate privacy setting before uploading unreleased client material. Generated images can still contain inaccurate text, inconsistent geometry, or unintended changes, and repeated references do not guarantee perfect continuity. Plan production time for selection and correction as well as generation.
It produces descriptive text. Use compatible image references when a character, object, or composition needs visual guidance, and inspect the resulting continuity.
No. Those features depend on the selected model. Check its capability labels and access requirements before planning a shot around a particular input.
Creative mode reconstructs detail. Precise mode is the more relevant starting point when fidelity matters, followed by a close comparison with the source.
It can generate and refine visual material. A layout editor is useful for exact, editable typography, dates, logos, and the final arrangement of production elements.