Text prompt to pixel
Every image starts as
static and a sentence.
This site is about the distance between those two things. What to write, what to leave out, which settings genuinely change the picture, and how to get from a promising first attempt to something worth keeping. Guides and reviews, written plainly, with the working shown.
Reviews & roundups
Which tool, for which job
Hands-on looks at the generators that do the work — single-tool reviews plus roundups for specific jobs. Written around the work, not the feature list.
Head to head
Comparisons
Two tools, side by side, on the questions that actually decide it — control, consistency, licensing, and how much fighting is involved.
Guides & reference
Learn the mechanism
The explainers, the how-to, and the reference. Start here if the words are working but the picture is not.
How we work
Editorial, and honest about it
Image generation attracts a lot of confident nonsense. Three commitments keep this site out of that pile.
We do not score things
You will not find a rating out of ten here. A number implies a measurement we did not take. We explain what a tool is good at and where it gets in the way, and let you decide.
Every result shows its work
When we describe an output, we say what produced it — the prompt, the settings, the number of attempts. A technique you cannot reproduce is not a technique.
Revisited when models move
Image models change quickly, and advice ages badly. When a guide stops matching what we see, we rewrite it rather than leaving it up for the traffic.
Some links to the tools we write about are affiliate links, and we may earn a commission if you sign up through one. That never buys a mention, a placement, or a kinder verdict — and it costs you nothing extra.
The pipeline
How a prompt becomes an image
Four stages sit between the sentence you type and the picture you get. Almost every practical question — why re-rolling helps, why one setting matters and another does not — is really a question about one of them.
Your words are encoded
The prompt is turned into a numeric representation of meaning. Phrasing, emphasis, and order all change that representation.
A field of noise is seeded
Generation starts from random static. The seed decides which static you get, which is why identical prompts can diverge.
The noise is guided away
Over successive passes the model subtracts what it believes is not your prompt. Guidance strength sets how hard it pulls.
The result is decoded
The finished representation is turned back into pixels, then usually upscaled and finished before anyone sees it.