Leonardo.ai

Realtime Canvas is a split-screen mode where you draw on one side and watch a finished image build on the other as you go. It's built for early brainstorming, when you're not sure what you want yet, rather than for final polish. Unlike a standard text box, it responds to strokes, so you can steer composition before committing to a prompt.

Phoenix and Lucid Origin and Realism are Leonardo's own models, tuned for prompt accuracy on complex, multi-element scenes. Phoenix handles scenes with several distinct objects, which is where older models often blend colors or mix up details. These first-party models also get relaxed, no-token generation on higher plans, which changes how much you can produce.

The Canvas gives you inpainting, outpainting, and compositing in one workspace. You can erase an unwanted object, extend a photo past its original edges, or blend generated pieces into an existing scene. This replaces jumping between an image editor and a generator for the same job.

Motion animates a still image into a short clip, usually a few seconds long, with camera moves like pan and zoom. It's enough for a teaser or a looping social post, not for long-form video. Premium third-party video models are also available in the same account, but they always draw tokens.

You can train your own model on your own images to lock in a specific character, product, or art style. Once trained, it generates consistently styled output that a generic model can't match. This is the feature that keeps a project visually coherent across dozens of assets.

The upscaler raises the resolution of generated or uploaded images and sharpens detail, so a small draft can become a print-ready file. Pro, Precise, and Creative upscale modes trade off between faithfulness and added detail. It runs inside the platform, so you don't need a separate upscaling tool.

The REST API exposes image and video generation, LoRA training, upscaling, and background removal for developers, billed pay-as-you-go. You create a key inside the web app, and webhooks can push results instead of forcing you to poll. This makes Leonardo usable as a backend engine, not just a website.