Glossary
Image generation
Image generation produces a new picture from a written description, rather than finding an existing one or editing a photograph you already had, which makes it a drafting tool before it is a production one.
In plain terms
You describe a picture and one appears. It is genuinely useful for the stage of work where you are trying to show somebody what you mean, and it is much harder to use for the stage where the result goes out under your name. The distance between those two stages is where nearly all the disappointment in this subject lives.
Why it matters
It collapses the cost of showing an idea, which changes how design conversations happen: a mood board that used to take an afternoon takes ten minutes, and options that were never worth commissioning now get looked at. What it does not collapse is the work of producing something you can actually publish, which involves consistency, rights and a level of control the technology gives you only partially. Knowing which of those two things you are buying prevents most of the frustration.
How it works
You supply a description and get back an image that matches it approximately. Small changes in wording produce large changes in output, which makes the process feel more like negotiation than instruction, and it is why people who use these tools well keep the descriptions that worked.
Consistency across a set is the hard part and the one that decides most commercial uses. Getting one striking picture is easy. Getting the same character, the same product or the same style across twelve images is considerably harder, and it is precisely what a campaign, a set of illustrations or a product catalogue requires.
Rights are the question to settle before anything else, and the answer lives in the provider's terms rather than in any general principle. Who owns the output, whether it may be used commercially, and what happens if a claim is made about it are all documented, they differ between providers and between plans, and they change. Read them for the plan you are actually on.
Likeness and trademark are separate risks from copyright and are easier to walk into. A generated image resembling a real person or carrying something recognisable belonging to somebody else creates a problem no licence from your provider addresses, because the provider was never the party with rights to grant.
Two jobs people expect one tool to do
Seen in the wild
Describe the same scene three times, changing one word each time, and keep the description that worked rather than the picture, which is the reusable part.
ChatGPTTry to produce four images of the same subject that genuinely look like a set, which is the test that separates a drafting tool from a production one.
Google Gemini
Common misconceptions
People assume
We can use whatever it produces.
In fact
That depends on your provider's terms, your plan, and what happens to be in the picture. Commercial use, ownership and indemnity are three separate questions with documented answers that differ between products. Settle them before a campaign rather than after, because the answer is not a general property of the technology.
People assume
Getting a good image means we can get twelve.
In fact
One striking result is straightforward and a consistent set is the genuine difficulty. Same subject, same style, same lighting across a series is where most commercial projects discover the limits, so it is the thing to test first rather than the thing to assume once a single image impresses.
Telling them apart
Image generation vs Diffusion model
Image generation
What you are doing: producing a picture from a description, with all the practical questions about rights, consistency and control that come with it.
How most of it is done: the particular kind of model that starts from noise and refines towards the described picture.
One is the capability you buy and the other is the mechanism underneath it. You can use the first competently without knowing the second, which is not true of most pairs on this roster.
Questions
- Can we use generated images commercially?
- Often, and it is a contractual question rather than a technical one. Check what your provider's terms say about ownership and commercial use for the plan you are on, and check whether any indemnity is offered if somebody makes a claim. Those answers differ between products and change, so establish them for your own account.
- Why can we not get a consistent set?
- Because each image is produced afresh from a description, and a description does not pin down everything that makes two pictures look related. Some products offer ways of holding a subject or style steady across a series, and how well those work is the single most useful thing to test if a set is what you need.
- What about the text inside images?
- Historically poor and much improved, and still worth checking every character rather than glancing at. Signage, labels and captions are where errors survive review, because a plausible-looking word in a small typeface reads as correct until somebody who needs it correct looks properly.
Key takeaways
- It produces a new picture from a description, which makes it a drafting tool before it is a production one.
- Small wording changes move the result a lot, so the description that worked is the thing worth keeping.
- Consistency across a set is the hard part and decides most commercial uses.
- Rights, commercial use and indemnity are documented, differ by provider and plan, and belong settled first.
- Likeness and trademark are separate risks that no provider licence addresses.
Tools that use this
- ChatGPT
Iterating on a description, where the wording is the reusable artefact.
- Google Gemini
The consistency test, which is what separates drafting from production.
Last checked July 2026