Explore the Work Find the Books

The Verse

5 — 7 — 5

Haiku

The contemplative core: a poem of the natural world, traditionally anchored by a season word and turned on a moment of sudden attention. One person, alone, awake to the world for the length of a breath.

Explore →

5 — 7 — 5

Senryu

Wears the same shape as haiku but turns the lens from nature to people. Where haiku reveres, senryu watches: wryly, sometimes mercilessly, always honestly. Funny the way the truth is funny — because it is exact.

Explore →

5 — 7 — 7

Katauta

The oldest and strangest of the four, literally a “half-song.” A fragment, one half of an exchange: a question waiting on its answer. A katauta hands you the other half and waits.

Explore →

5 — 7 — 5

Zappai

The laugh. Historically the catch-all term for light, comic, miscellaneous verse — the popular, vernacular cousin the purists looked down on, in no small part because ordinary people loved it.

Explore →

The Form

On the Art of the Poem-Painting, and One Poet’s Chronicle of Its Newest Form

I. The History of Haiga

The first illustrated haiku appeared in the work of Nonoguchi Ryūho in the seventeenth century, but it was Bashō who established the form as a serious artistic practice and, more importantly, gave it its philosophical foundation. For Bashō, poetry and painting were not separate arts but two expressions of the same impulse: the impulse to stop time, to hold a moment of perception so precisely that it could be entered again by anyone who encountered it.

Bashō’s haiga were characteristically simple. He was a poet who painted, not a painter who wrote — his brushwork was spare, his figures gestural, his landscapes suggestive rather than descriptive. This restraint was not a limitation but a principle. The traditional aesthetic of haiga, as the scholar Leon Zolbrod has written, is characterized by “free and flowing line work and elimination of unnecessary detail.” The paintings often carry “a light or frivolous touch suggestive of irony or amusement, even when the subject of the painting is serious.” Simplicity and irony: the twin virtues of the form.

The master who brought haiga to its greatest artistic height was Yosa Buson (1716–1784), widely considered the second of the great haiku masters after Bashō, and the only artist in Japanese history to be ranked among the greatest in both poetry and painting. Where Bashō’s haiga were a poet’s sketches, Buson’s were a painter’s poems: richly detailed, luminous, technically accomplished.

What Buson understood, and what his haiga demonstrate with extraordinary clarity, is that the poem and the painting in a haiga are not required to match. They are required to resonate. “Both Buson and Bashō considered painting and poetry as two forms of the same activity, the same ideal,” one scholar has written — not the same subject, not the same image, but the same reaching toward what cannot be fully said in either medium alone.

What Buson understood, and what his haiga demonstrate with extraordinary clarity, is that the poem and the painting in a haiga are not required to match. They are required to resonate. “Both Buson and Bashō considered painting and poetry as two forms of the same activity, the same ideal,” one scholar has written — not the same subject, not the same image, but the same reaching toward what cannot be fully said in either medium alone.

After Buson, haiga continued as a practice among the Edo period aesthetes — composing haiku and painting accompanying pictures was a common pastime at friendly gatherings, a communal entertainment as well as a serious art. The great novelist Ihara Saikaku was among the many who participated. The form was never confined to professional artists; it belonged to anyone who could hold a brush and attend to the world.

In the twentieth century, as haiku traveled westward and found new practitioners in English and other languages, haiga traveled with it — adapted for photography, for digital illustration, for collage and mixed media. The Academy of American Poets has noted that contemporary haiga artists “continue to combine poetry and image” in “traditional or modernist ways,” painted on parchment or “assembled in Adobe Photoshop and lined up on a webpage.” The form is as flexible as its name is ancient.

This collection is the newest point on that long line — and also something the tradition has not produced before: a record of a machine learning to see.

II. The Polaroid Dimension

On November 26, 1948, the first Polaroid Land Camera went on sale at Jordan Marsh department store in Boston. It sold out the same day. What Edwin Land had invented was not merely a faster camera — it was a different relationship between the moment of capture and the image that resulted from it. Before Polaroid, photography required time: film to be developed, prints to be made, days or weeks between the shutter click and the photograph in hand. Polaroid collapsed that gap to seconds. The image emerged from the camera while you were still in the moment that produced it. The Polaroid was, Land understood, an act of instant memory.

Some years ago, on a trip to London, I came across Polaroid cameras and film — the brand had been revived after its 2007 bankruptcy, and the film was available again. I purchased some and brought it home. I also acquired a Polaroid printer, a device that allows photographs taken on a smartphone to be printed in the distinctive Polaroid format: the square image, the white border, the slight chemical imperfection that no digital filter has ever quite replicated.

The Polaroid camera itself, I discovered, still pretty much disappointed. The film was expensive, the results unpredictable, the charm considerable but the practicality limited. The printer was another matter. With the printer I could bring the Polaroid aesthetic — its distinctive frame, its quality of captured-in-the-moment, its sense of being a single unrepeatable image rather than one of a thousand identical digital files — to photographs made on the phone.

When I began making haiga, the Polaroid frame presented itself immediately as the right format. The reasons were deeper than aesthetics. The haiku moment — the instant of perception that the form was built to capture — has always been about the present tense, the thing that is happening right now and will not happen again in exactly this way. The Polaroid is the photographic form that most honestly represents that quality. It is an image made in a moment, fixed in that moment, slightly imperfect in the way that all moments are imperfect. It does not pretend to be timeless. It is, very specifically, then — and it knows it.

The Polaroid format also does something formally important for the haiga: it creates a defined space in which the poem and image coexist. The white border contains them. The square frame equalizes them. Neither the poem nor the image is permitted to overwhelm the other. They share the space the way a conversation shares a room — each present, each distinct, each aware of the other.

There is something additionally right about the Polaroid for the senryu and the zappai specifically. The Polaroid has always had a slightly comic quality — its imperfections, its unpredictability, the way the image emerges gradually as if arriving at a decision about itself. It is not a solemn medium. It is an intimate one. These qualities suit the human-centered, often humorous, always particular world of senryu and zappai in a way that a formally composed studio photograph never could.

III. MidJourney — An Engine for the Imagination

In February 2022, David Holz — a physicist and technologist who had previously built hand-tracking hardware at Leap Motion — launched a new kind of research laboratory in San Francisco. He called it Midjourney. Its purpose, in Holz’s own words, was to expand “the imaginative powers of the human species.” The goal, as he described it, was not to make imaginative machines but to make humans more imaginative — “an engine for the imagination,” accessible to anyone who could type a description of what they wanted to see.

I joined Midjourney in August 2022, when the platform was still in early public beta. I have used it continuously since, through every major version of the model, making haiga — seventeen-syllable poems paired with AI-generated images in the Polaroid format — as both an artistic practice and, as it has turned out, an inadvertent chronicle of how machine vision developed in real time.

What Holz built was not simply a tool. It was, in his conception, a community — a social space for making beautiful things together, governed by a genuine aesthetic ambition. Midjourney’s community of users, which grew to millions within the first year, was characterized from the beginning by a shared commitment to artistic quality, experimentation, and mutual appreciation. The Discord server through which Midjourney operated was, at its best, something like the Edo period gatherings where poets and painters shared their work: a communal pursuit of beauty in real time, each participant seeing what the others made and responding with their own attempts.

IV. A Chronicle of Machine Vision — The Seven Versions

The haiga in this collection were made across Midjourney versions 1 through 7, spanning the entire publicly accessible history of the platform. They are not merely a collection of poems and images — they are a document of how artificial intelligence learned to see.

Volumes I–III: Versions 1, 2, and 3 — The earliest Midjourney versions produced images that were distinctly and unmistakably machine-made: dreamlike, impressionistic, often anatomically inaccurate in ways that were simultaneously strange and compelling. Faces dissolved at the edges. Hands had too many fingers. Backgrounds shimmered with a quality that had no analog in traditional photography or painting but had a great deal in common with certain kinds of psychedelic art. These versions were ethereal, arty, and gloriously unreliable. They were also, in their way, the most purely surrealist images in the collection — machines generating images the way fever generates dreams, with an internal logic that was not always visible to the viewer but was always present.

Volume IV: Version 4 — This was the version that changed everything. MidJourney 4 produced the first images that looked, to a casual eye, almost photographic. Faces were recognizable. Spaces were coherent. Light fell in ways that obeyed physical laws. The AI had achieved something like verisimilitude — and the AI art community responded with a mixture of amazement and unease that has not entirely settled since. For the haiga, version 4 introduced a new challenge: how to make images that were genuinely interesting rather than merely impressive. A photographic image paired with a poem risks becoming illustration. I had to learn to direct the model away from the obvious.

Volumes V and VI: Versions 5, 5.1, 5.2, and 6 — The middle versions of Midjourney are the most technically refined and, in some respects, the most challenging to use well. They can produce extraordinary beauty — imagery of a quality that would have taken a skilled illustrator days to produce — but they tend toward a kind of perfection that is at odds with the haiga aesthetic. The challenge across these versions was to resist the pull toward the polished and the obvious, to find images that had the quality of a well-made sketch rather than a finished commercial illustration. This required learning the model’s tendencies and working against them — asking for minimalism, for abstraction, for styles and genres that pushed against the photorealistic default. Sumi-e ink painting. Cartoons. Macro photography. The model’s resistance to these requests was itself instructesistance to these requests was itself instructive.

Volume VII: Version 7 — The most recent stable version of Midjourney brings new capabilities and a different relationship between prompt and output. The model’s understanding of style, context, and visual metaphor has deepened considerably. It has begun to produce images that feel less like responses to descriptions and more like interpretations — less “here is what you asked for” and more “here is what I made of what you said.” For the haiga poet, this shift is significant. The AI has begun to function less like a search engine and more like a collaborator with its own aesthetic instincts. Whether those instincts are trustworthy, surprising, or both depends entirely on the poem.

I made a decision early in this project that I have never abandoned: I do not go back. Earlier versions of Midjourney remain accessible, and their distinctive qualities are often tempting. A poem that might have produced a more interesting image in version 3’s dream logic is now made with version 7’s interpretive sophistication. The earlier images remain as they were made, in their version, with that version’s understanding of the world. This is not nostalgia prevention — it is fidelity to the chronicle. Each haiga is a record of the moment it was made, in the version of machine vision that existed at that moment. Going back would falsify the record.

Version 8 of Midjourney was in beta at the time of this writing. The next collection will be made with it. The chronicle continues.

V. The Process — How a Haiga Is Made

My process for making a haiga begins, as all poetry begins, with a line. An image, an idea, something heard or read or observed that demands to be held in seventeen syllables. The poem comes first, written in the discipline of strict 5-7-5, shaped until it is as precise and compressed as the form requires.

Once the poem is complete, it becomes the prompt. I enter the text of the poem directly into Midjourney using the /imagine command — the same words, in the same order, exactly as they appear on the page — and receive four candidate images in response. I then typically run additional rounds using the same poem with style modifiers appended: “in the style of minimalism,” “sumi-e ink painting,” “cartoon,” “macro photography.” Each modifier produces four more candidates, each a different interpretation of the same seventeen syllables.

The selection criterion is deliberately subjective and deliberately demanding: the image must move me. It must produce a reaction — laughter, sadness, discomfort, surprise, recognition — before I consider whether it relates to the poem at all. An image that fails to stand alone on its own merits fails regardless of how cleverly it illustrates the poem. Illustration is not the goal. The goal is resonance.

I look for the unexpected. If three of four images show a dog and one shows a bird, I consider the bird — the three-of-four response indicates that the model found the dog obvious, and the obvious is rarely interesting. I do not use the variation tools or inpainting functions to edit or modify images. I work with what the model produces or I start again. This constraint is not arbitrary: it preserves the collaborative nature of the process. The model makes what it makes. The poet chooses or refuses. The choice, when it comes, is almost always a surprise.

Occasionally the process reverses. An image appears — in a prompt round, perhaps, that produced nothing suitable for its original poem — that demands a poem of its own. In these cases I begin with the image and write toward it, finding the seventeen syllables that complete what the machine began. The haiga that result from this reversed process are among the most interesting in the collection, because they represent a genuine collaboration: the machine making the first move, the poet responding.

VI. Haiga for the Modern Age

There is a question that surrounds AI-generated art, one I have heard many times and expect to hear many more: is it art? The question is usually meant to challenge the legitimacy of the AI’s contribution — to assert that because a machine made the image, the image lacks something essential that only human hands can provide.

My answer is to point to the haiga tradition. From Bashō’s spare ink sketches to Buson’s luminous paintings, haiga has always accommodated a range of relationships between the poet and the image-maker. Some of the greatest haiga in the tradition were collaborations — Bashō’s poetry paired with Morikawa Kyoroku’s paintings, each artist responding to the other’s work rather than producing a unified creation from a single sensibility. The question of who made which part was never the point. The question was always whether the combination produced something that neither element could produce alone.

Midjourney is a new kind of collaborator. It has no sensibility in the human sense — no lived experience, no emotional history, no memory of a childhood Polaroid camera found in a London shop. But it has processed an extraordinary range of human visual expression, and from that processing it produces responses to language that are genuinely surprising, genuinely varied, and occasionally genuinely beautiful. It is, in David Holz’s phrase, an engine for the imagination — not the imagination itself, but something that ignites it.

The haiga form is, in this sense, perfectly suited to the age of AI image generation. It has never required that the poet and the image-maker be the same person. It has never required that the image illustrate the poem or that the poem explain the image. It has required only that the two elements, placed together, produce a third thing — a resonance, a question, a moment of recognition — that justifies the combination. By that standard, these haiga stand or fall on their own terms, as haiga have always done.

There is something additionally fitting about the Polaroid format in this context. The Polaroid was itself a technology that changed what it meant to make an image — that collapsed the distance between capture and result, that put the process in the hands of anyone rather than the professional alone, that made photography immediate and personal in ways it had never been before. Edwin Land’s invention was contested too, in its time. Some argued that instant photography was not photography as serious practitioners understood it. The argument did not survive contact with what people actually did with the cameras.

So my haiga are two things at once: the newest entry in a four-hundred-year-old form, and a record of a very new machine in the first years it learned to make pictures. Poems I counted by hand, set beside images that surprised me. The oldest of forms, remade once more, in the newest way there is.