For a long time the standing joke about image generation was that the models could produce a photorealistic dragon and could not spell. Any request involving readable text produced letter shaped noise, which made the entire category useless for the large amount of design work that involves words. Ideogram made text rendering its focus and largely solved the problem, which is a genuine technical achievement and the reason to pay attention to it. The text rendering is the whole proposition and it delivers.
Short phrases, logos, poster headlines and sign text come out readable and correctly spelled at a rate that the general purpose models still do not match. If your work involves words inside images, that is not a marginal improvement, it is the difference between a usable tool and one you cannot use at all. Anyone who has tried to produce a mock poster with a general model and given up after twenty attempts will recognise what this changes. The prompt guidance for typography is the strongest part of the learning material.
How to specify the text you want, how to describe the style and placement, how to separate the words from the surrounding scene description. These are specific and useful and they reflect the tool's actual behaviour rather than generic advice about being descriptive. Given how thin the rest of the material is, this section stands out as something that was clearly written with care. The style controls are simpler than the alternatives and I count that as a virtue.
Rather than a long list of parameters requiring experimentation to understand, there are style presets and a smaller set of adjustments that behave predictably. That reduces the learning curve for people whose interest is design rather than tooling, and the trade against fine grained control is a reasonable one for the intended audience. The free tier allows meaningful daily generation, which matters for learning. Understanding how a model responds to prompt variations requires many attempts, and a tool that limits you to a handful teaches nothing.
Here you can iterate enough to build genuine intuition without paying, and that is the right approach for a tool where competence comes from repetition. The official material is sparse and that is the main criticism. There is a help centre, some prompt guidance and a gallery of examples, and it does not go deep. The real knowledge about what works lives in community spaces where people share prompts and results.
That pattern is universal in generative image tools and it does mean the vendor is leaving the education to its users, which works and is not ideal. General image quality outside the speciality is where honesty is needed. For photorealism, complex scenes and artistic range, the leading general purpose models are ahead. This tool is excellent at its specific thing and merely good at everything else, and the learning material naturally does not tell you that.
My advice is to use it for text heavy work and something else for general imagery, which is what most people doing this seriously end up doing anyway. Longer text remains unreliable. Short phrases work well, a full paragraph does not, and the failure rate rises with the length of the string. The material does not set that expectation clearly, and someone attempting a text heavy layout will waste generations discovering the boundary.
Knowing where it sits saves time, and the practical rule is that headline length is safe and body copy is not. Licensing deserves a proper read. What you can do commercially with generated images, how that varies between free and paid tiers, and what rights you hold are all questions with real answers that people routinely skip. Anyone using this for client work should read the current terms rather than assuming they match another tool's.
Three point seven for a tool with a real and well executed differentiator, supported by learning material that covers the speciality reasonably and leaves everything else to the community. The text rendering is genuinely the best available and the surrounding education does not do enough to help people exploit it.