Leonardo sits at a useful point between the tools that give you a prompt box and the ones that expect you to run your own inference. There is more control here than most consumer products offer, and the learning material does a decent job of explaining what the controls do. The custom model training material is the part worth the time. Training a model on your own images so that outputs share a consistent look is what turns generation from a novelty into something usable for a project, and the guidance is realistic about what a good training set requires and about the fact that a poor one produces a poor model.
The control features are documented properly. Canvas editing, reference images, negative prompts and the various guidance settings are the difference between accepting whatever you get and directing towards something specific, and the material treats them as a system rather than as a list of toggles. That is the right approach and it is where the ceiling of the tool sits. The daily free token allowance deserves a mention because it changes how people learn.
Being able to come back every day and experiment without paying means you build intuition over weeks rather than burning a trial in an afternoon, and more products should do this. The gap is the same one every image generation resource has, which is that the material teaches the controls and not the judgement. Why one image looks professional and another looks like generated filler is mostly composition, colour relationships, restraint and knowing when to stop, and none of that appears anywhere. A user with visual training gets a great deal from this tool.
A user without one produces a very large quantity of mediocre images and concludes that they need a better prompt. Documentation lag is the usual problem. Features ship faster than the writing, the help centre describes interfaces that have moved, and the community fills the gap unevenly and with the usual survivorship bias in what gets posted. Token consumption during iteration is the practical trap.
Exploration costs the same as production, most generations get discarded, and people budget for the images they keep rather than for the ones they made on the way. Anyone planning a project should assume a discard ratio and price accordingly. Commercial rights guidance is thinner than the audience warrants. A lot of these users are producing work for clients or for sale, and they need a clearer answer about what they own and what exposure they carry than terms of service provide.
It is the standard omission across this category and it does not stop being a problem for being common. Three point five. Approachable material for a genuinely capable tool, strongest on custom models and control, weakest on everything that depends on taste, and quiet on the commercial questions its users actually face.