Images: brief the AI like an art director

Yesterday you learned to think before you make. Today that rule meets pictures: you'll describe an image the way an art director briefs an illustrator — five decisions, one sentence — and walk away with a card, a flyer or a wallpaper you'd actually use. Everything on free plans, as always.

1Where the picture comes from — and why your words decide everything

Here's what actually happens when you type a sentence and an image appears, in one honest paragraph. These systems learned from millions of pairs of images and their written descriptions. Over enough examples, they learned what words look like: what "golden retriever" does to shapes and fur, what "watercolor" does to edges, what "at dusk" does to light. When you type a description, the model renders a brand-new image that matches your words — it isn't searching the web for a photo, and it isn't cutting and pasting from existing pictures. It's drawing what you described, from everything it learned about how descriptions map to images.

That mechanism has one consequence that explains every disappointing AI image you've ever seen: the model renders your description — and decides everything you didn't describe. Ambiguity isn't an error; it's a delegation. Type "a dog in a park" and you've delegated the breed, the season, the time of day, the camera angle, the art style and the mood to a machine that will pick the most statistically average option for each. The result is technically correct and completely generic — a stock photo of nobody's dog in nobody's park.

If this sounds familiar, it should. It's the Briefing Formula from Day 2 all over again: whatever you don't specify, the assistant guesses, and guesses are generic. And it's Think → Make from Day 6 in new clothing: the image generator is the builder, not the architect. The thinking — what exactly should this picture show, in what style, for what use — is your job. Today's five-part structure is how you do that thinking for pictures.

2The Image Brief: five decisions that make the picture yours

An art director never tells an illustrator "draw something nice." They hand over a brief. Yours has five parts — this is the Image Brief, and it covers every decision the model would otherwise make without you:

• Subject — who or what the picture is of. The main noun plus one or two specifics: "a golden retriever puppy in a blue party hat", "a steaming mug", "a poster with the text 'GARAGE SALE'".
• Setting — where and when. Place, time of day, weather: "in a pile of autumn leaves in a city park at sunrise", "on a wooden desk by a rain-streaked window".
• Style — how it's made. "Photorealistic", "children's-book watercolor", "flat illustration", "vintage travel poster", "minimalist line drawing".
• Mood — what it should feel like. "Warm and nostalgic", "playful and bright", "calm and quiet". Mood quietly drives the light and the colors.
• Format — what shape, for what use. "Square, for a birthday card", "wide landscape, for a desktop wallpaper", "vertical poster". Naming the use helps the model compose for it.

Put all five in one plain sentence and you get something an art director would recognize:

Two reassurances before you worry about getting this "right". First, the five parts are a checklist, not a straitjacket — a plain sentence that touches all five beats any keyword soup, and the order doesn't matter. Second, you don't need art vocabulary. "Looks like a kids' picture book" is a perfectly good style. "Cozy, like a Sunday morning" is a perfectly good mood. The model learned from human descriptions; ordinary human language is exactly what it understands best.

3The second superpower: refine in the same thread

The first image is a first draft — the same rule you learned about text on Day 1. But here's what most beginners never discover: you don't have to start over. Because you're working in a conversation, the thread remembers the picture it made, and you can direct changes in plain language: Same image, but at dusk. Keep everything, make the mug red. Remove the text on the wall. Each reply is an edit instruction to an artist who never gets tired of revisions — not a fresh pull on a lever.

This is the difference between regenerating and refining, and it's worth saying precisely: every regeneration with the same words is a coin flip; every changed word is a decision. If the image is wrong, something in the brief was missing or unclear — name it. "Make it feel warmer" beats trying again. "Seen from across the street" beats hoping. You are the art director; the review round is where your taste enters the picture.

One honest expectation-setting note: hands, faces and small lettering keep improving, but they're still where image models most often slip. If a detail matters — the number of fingers, the spelling on a sign — look at it closely before you use the image, and name the fix in your next refinement. The Confident Intern from Day 1 draws pictures too.

4Three tools, three jobs: who paints, who letters, who writes the brief

This is the staff model from Day 3 doing real work. The big three split today's job cleanly:

• Gemini — the volume tool. The most generous free image generation of the big three by a wide margin — a daily free allowance big enough to treat as practically unlimited for practice (the exact cap changes often; check the current limits inside the app). That makes it your practice studio — the place to learn briefing by doing, where a wasted image costs you nothing.
• ChatGPT — the precision tool. Its image generator is the strongest at clean, correctly spelled text inside the image — posters, invitations, signs. But the free image allowance is small and changes often (check the current limits inside the app), so spend each one deliberately: brief first, generate second.
• Claude — the brief-writer. Claude generates no images at all. But it's the strongest writer of the three — which makes it a superb image-brief factory: describe what you need, get five art-directed briefs in five styles, carry the best one to Gemini.

1. Open gemini.google.com and sign in with your Day 1 account. Look at the bottom of the page for the message box. an empty message box at the bottom, your past chats in the sidebar.
2. Copy-paste this full Image Brief into the message box and send it. Generating an image takes longer than text — give it a few seconds.
   Copy-paste prompt · GeminiCreate an image: a cozy home office in Scandinavian style, morning light through a large window, a steaming mug on the desk, soft pastel colors, photorealistic, landscape format.
   the image appears right in the chat, like a photo in a messenger. Notice how much of it you recognize from the brief — the light, the mug, the colors.
3. Now refine — in the same thread, not a new chat. The conversation is what remembers the picture.
   Copy-paste prompt · Gemini (same thread)Same scene, but evening, warm lamp light, and add a sleeping cat on the chair.
   the same room, re-rendered: lamp light instead of morning sun, a cat on the chair. The scene survived; only what you named changed.
4. Make one more refinement of your own — change exactly one thing ("make the walls deep green", "seen from the doorway"). This is the habit: change words, not dice. the one thing you named changes while the rest mostly holds. That's directing, not gambling.
5. Save your favorite: hover over (or tap) the image and use the download icon. the image file lands in your downloads folder, ready to use anywhere.

1. Decide the exact text your image must carry, word for word, before spending a generation. Write it down. (This is Think → Make at miniature scale.) a one-line note with your real text, e.g. "GARAGE SALE · SATURDAY 10AM".
2. Open chatgpt.com, start a new chat, and paste the brief with your text in the quotes:
   Copy-paste prompt · ChatGPTCreate a simple poster: the text "[your real event — e.g. GARAGE SALE · SATURDAY 10AM]" in bold, friendly lettering, bright illustrated style, sunny suburban background, square format.
   a poster with your words rendered in actual legible lettering — the thing image tools used to be famously bad at.
3. Proofread the image the way you'd proofread an email: read every word on it, out loud if you can. The lettering is usually right — "usually" isn't "always". If anything is off, refine in the same thread:
   Copy-paste prompt · ChatGPT (same thread)Same poster, keep everything — but fix the text so it reads exactly: [your text, spelled the way you want it].
   the corrected poster. Check it again — text earns a second read.
4. Download it the same way as before (hover or tap, then the download icon). a square poster file you could genuinely print or post today.

1. Open claude.ai, start a new chat, and order five briefs at once:
   Copy-paste prompt · ClaudeWrite 5 image-generation prompts for [your real need — e.g. a birthday card for a 60-year-old jazz fan]. One sentence each, and each should cover subject, setting, style, mood and format. Vary the style: photorealistic, watercolor, flat illustration, vintage poster, minimalist.
   five numbered one-sentence briefs, each a complete Image Brief in a different style — a stylistic menu for the same idea.
2. Pick your favorite and iterate on it like any draft (Day 2 habit): Make #2 warmer and more playful, and end it with: square format, for a printed card. the brief sharpened to your taste — before a single image generation was spent anywhere.
3. Copy the finished brief, paste it into Gemini, and generate. an image built from a professionally written brief you art-directed in about a minute — the staff model from Day 3, working as a team.