AI images · one line in, one image out

AI Images: What One Sentence Makes

The sentence was short. The light did the rest. Nine real renders, each quoted with the line that made it.

Close portrait of Luffy grinning at sunset, straw hat pushed back

Close portrait of Hori laughing in a school corridor, ribbon loosened

Close portrait of Tanjiro in falling snow, scar and earrings in frame

Luffy on the bow of the Going Merry at sunrise, arms open, straw hat raised, sea spray below — **Monkey D. Luffy** — Wide shot, low sun. The line named the ship and the spray; the sunrise set the temperature.

Nami leaning over a chart table in a harbour tavern at night, lit by a single lantern — **Nami** — Interior, one lantern. Everything outside the lantern pool falls into blue.

Zoro on a wet stone bridge in the rain, one sword drawn, grey dawn behind him — **Roronoa Zoro** — Grey dawn, rain. One wet highlight along the blade is the whole focal point.

Raiden Shogun descending the stone steps of Tenshukaku under branching lightning — **Raiden Shogun** — Storm key light. The shrine stays in silhouette so she reads first.

Zhongli at a Liyue teahouse window at night, lanterns and harbour lights outside — **Zhongli** — Lantern rain outside, cool room inside: two lights, one subject.

Nahida reading in a dark Akademiya archive, a single sunbeam falling across the shelves — **Nahida** — One hard shaft of noon light; the room is allowed to stay dark.

Hori at the classroom window after the last bell, cherry petals blowing past the glass — **Hori Kyouko** — Backlit window, warm on the face, cool across the empty desks.

Miyamura alone on the school roof at dusk, city lights coming on below — **Miyamura Izumi** — Blue hour with one warm window behind him.

Tanjiro crossing a snowy mountain pass at night with a lantern, footprints behind him — **Tanjiro Kamado** — Lantern key in a blue snowfield: the only warm point in the frame.

A sentence, then a world that remembers it

A prompt box hands you a picture and forgets it. In Studio the picture lands as a work with its own page, the character is a material with a face you can point at, and the next image starts from that world instead of from zero — which is why the ninth render is as steady as the first.

The character keeps its face — A character is a material with a cover image. That image goes in as the reference, so the face, hair and outfit survive the change of scene.
One look per world — Each world carries its own visual style. Every image follows it without you retyping the style sentence.
Every render is a work — Each result gets a page and a URL: reopen it, compare it with the next take, or put it on a page like this one.

Questions before your first line

What do I actually control?

The scene, the framing, the light, the style words, and one or more reference images. What you do not control is the model's drawing hand: it decides the detail inside the frame you described.

How do I keep the same face in the next image?

Attach the first image, or use the character's own cover, and describe only what changes — the outfit, the light, the pose. A reference beats a longer description every time.

Which format should I choose?

3:2 for scenes and posters, 3:4 for portraits and phone screens, 1:1 for cards and avatars. Decide before you generate: changing the frame later crops the composition.

Can I put text inside the image?

Short, large words usually survive. Paragraphs and long titles do not, so generate the picture clean and set the type in an editor afterwards.

Why do hands and eyes go wrong?

They are the smallest shapes in the frame and they get the fewest pixels. Keep faces and hands large and near the centre, or crop the frame afterwards.

Can I use a photo of myself as a reference?

Yes, your own photo is the normal way to keep a likeness. Someone else's face needs their permission, and a real person's likeness is not what this is for.

Are these nine images from a real session?

Yes. Each was generated with the composer prompt on this page and the line quoted under it is the input that produced it.

Can I sell what comes out?

For characters and worlds you do not own, treat the result as fan work: fine to share, not to sell. Your own characters and places are yours to use.

One sentence is the brief.

A character, a place, one light. The rest is the render.

Make My Image