Recraft aims at a narrower audience than most image generators and is better for it. The assumption is that you are producing assets for a brand or a product rather than making one striking picture, which means the problems that matter are consistency across a set, editability of the output and getting something that fits into a design workflow. The learning material is organised around those problems, and after a lot of generic prompt guides elsewhere, that focus is a relief. Vector output is the headline capability and the guidance treats it fairly.
Generating something you can open in a vector editor and adjust is genuinely different from generating a raster image you can only regenerate, and for icons, simple illustrations and logo adjacent work it changes the workflow. The material is reasonably honest that the paths are not always clean, that complex illustrations produce messy geometry, and that you should expect to tidy up. It still undersells the tidying. Anyone expecting hand drawn path quality will spend longer in the editor than the tutorials suggest.
Style consistency gets the attention it deserves. Producing forty images that look like they belong together is the actual requirement in design work, and it is much harder than producing one good image. The style training material walks through building a coherent reference set and is realistic that a sloppy set gives you a muddled style. It also explains applying a style across a batch, which is the part that saves real time.
This is the strongest section of the resource. Range is the honest trade. For photographic realism, unusual compositions or anything where you want breadth rather than repeatability, general purpose models are better, and the material naturally does not draw that comparison. Knowing where the ceiling sits saves you from fighting the tool on work it was not built for.
Credit consumption during style development is the practical trap. Iterating on a brand style means many generations that exist only to inform the next attempt, and that exploratory phase costs more than the final production run. The pricing is documented and the pattern of consumption is not, so budget for the iteration rather than for the output. Licensing is the omission that matters most for the target audience.
A designer producing commercial work needs a clear answer on rights, on what the model was trained on and on the exposure they carry when a client asks. The material offers terms of service and no practical guidance, which leaves the professional user carrying a risk they may not have priced. It is the standard failing of this category and it stings more here because the audience is explicitly commercial. Three point eight.
Focused, sensible material for a tool that solves the consistency problem better than its competitors, marked down for underplaying the vector cleanup, for quiet on iteration cost and for leaving commercial designers to work out the licensing on their own.