Say It, See It: The Visual Dictionary of AI Image Prompts (2026 Edition)

A picture book that teaches you to make pictures. 108 pages, 200 AI pictures, and beside every one the exact words that made it. For ChatGPT, Gemini, Grok and any other picture tool. 2026 edition, tools checked October 2026.

Most people type a few words into an AI picture tool, get something impressive that is not what they wanted, and do not know what to change. The answer is nearly always a word they did not know to use: golden hour, low angle, negative space, gouache, 35mm film. This book shows you those words. Each one has its own picture, so you can see what a word does before you use it.

How it works

The heart of the book is a dictionary of 114 looks in eleven chapters. Each chapter takes one subject and changes one thing about it:

• Light: one woman, twelve kinds of light, from golden hour to direct flash.
• Weather and hour: one stone cottage at dawn, in storm, in snow, under stars.
• Framing and lens: one fisherman, from wide shot to extreme close-up.
• Composition: one red sailboat, placed eight ways.
• Kinds of photograph: product, food, macro, aerial, street, wildlife and six more.
• Film and finish: one seaside kiosk as digital, 35mm, black and white, instant film, a 1970s snapshot.
• Color: one bicycle and one door in eight color schemes.
• Mood: one armchair, six moods.
• Medium: one fox, twenty-four ways: watercolor, linocut, stained glass, pixel art, embroidery.
• Material and Period: a pear made of glass, chrome, wool and ice; a corner café from a medieval manuscript to the 1980s.

Because only the words change from plate to plate, you can see exactly which words did the work. They are marked in yellow, ready to lift into your own prompt. Every chapter opens with a picture index of its plates and one plate printed across a whole page.

What else is inside

• The method (7 pages): one picture built from the single word "dog" to a finished watercolor, one sentence at a time, then the same prompt sent four times so you can see what holds and what changes.
• Same words, nine tools (6 pages): three prompts sent unchanged to nine models from OpenAI, Google, xAI, Black Forest Labs, Ideogram and Recraft, with every result printed, including the misses.
• Changing a picture (9 pages): change one thing and keep the rest, redraw a photo as a painting, keep a character the same from scene to scene, get lettering spelled right, thirteen sentences to send when a picture comes out wrong, and two pages of real faults shown beside their fixes.
• 24 recipes (25 pages): finished prompts for a headshot, a product shot, food, a room, a pet portrait, a picture-book page, a travel poster, a book cover, a logo, stickers, a pattern, a coloring page, an invitation, an infographic, icons and more, each with its picture and with the words you swap marked.
• The tools, checked October 2026 (8 pages): what the picture feature is called in ChatGPT, Gemini, Grok, Microsoft Copilot and Midjourney, where you start, how you change a picture and what marks it leaves, plus Ideogram, Recraft, FLUX and Adobe Firefly. Then what each maker's terms say about who may use your pictures, in the makers' own words, and a checklist to run before you publish.
• An index of every look, and the list of sources.

Every picture is real

Each plate was made on 9 or 10 October 2026 by the tool named on its page, from the words printed beside it. Nothing was retouched. Each prompt was sent once and what came back was printed, with the exceptions the book sets out on page 5: one prompt was sent four times on purpose, to show how results vary; four plates were made a second time, larger, to fill a page; and fourteen first pictures were set aside because of a slip in a hand, a face or a background figure. Those fourteen faults are shown on two pages beside the plates that replaced them, so you learn what to look for. Every plate where a tool did not do what the words said is listed, with what went wrong.

Every fact is checked

AI tools change month to month, and most guides are out of date when you buy them. Every statement in this book about a tool comes from that maker's own pages, read on 7 and 9 October 2026: 85 official pages, 173 sentences, each checked word for word against a saved copy. Where a maker's pages are silent, the book says "not stated" and does not guess.

What you download

• The book as a PDF (108 landscape pages, made to be read on a laptop, tablet or a phone turned sideways; it prints on Letter or A4 paper set to landscape).
• A ZIP with every prompt in the book in one plain-text file, by plate number, ready to copy and paste, and the list of sources.

Good to know

• Independent guide. Not affiliated with or endorsed by OpenAI, Google, xAI, Microsoft, Midjourney, Ideogram, Recraft, Black Forest Labs, Adobe or Higgsfield.
• You need a picture tool, which is not included. ChatGPT, Gemini and Microsoft Copilot can all be used without a paid plan, with limits, according to their makers' pages.
• A picture tool gives a different picture every time, so your pictures will not match the ones printed. The book shows this on page 11.
• The plates were made by sending each prompt straight to the model through the Higgsfield and Recraft services, not by typing into the chat apps. The models are the ones the makers publish under these names.
• No plate was made with Midjourney, Copilot or Adobe Firefly. Those tools are described from their makers' pages only.
• No prices and no picture limits are printed, because they change too often.
• The pages on who may use a picture are not legal advice.
• The book uses the spelling "color", because that is the word the tools use.