AI Can Now Build Your Whole Campaign Visual Set. The Skill Is in the Brief.

The question "can AI make marketing images?" was settled months ago. The more useful question is whether you can write a brief that tells it exactly what to make — and get back something coherent enough to actually deploy. That's the bottleneck, and it got a lot smaller this week.

What shipped and why it matters

OpenAI updated ChatGPT's image generation on April 21 to use reasoning rather than pattern-matching alone (OpenAI blog, Apr 21; TLDR AI, Apr 22). The practical difference: it can now generate up to 8 coherent images from a single prompt, maintain consistent visual style across a set, render text accurately — product names, taglines, button copy — and output at 2K resolution. Standard mode is free. A thinking mode for paid users handles more complex multi-element scenes.

What this actually changes is not the quality of a single image. It's the ability to generate a visual set in one session: three social card variations, a banner, a landing page hero, a story format. If the brief is right, they look like they belong together.

What to put in a visual brief

A prompt that produces one acceptable image and a brief that produces eight usable campaign assets are not the same thing. A visual brief for AI has six components:

• Subject and product: What is shown and what it actually looks like — shape, color, materials, scale
• Placement: Where the image will appear (Instagram story, email header, LinkedIn banner) — this determines dimensions and focal point
• Audience and context: Who sees this and in what mental state — someone scrolling fast, someone mid-decision
• Tone words plus one negative: What the visual feels like, and one clear exclusion to prevent the default (elegant, direct, warm — not corporate-polished)
• Text that must appear verbatim: Exact product name, tagline, any required copy
• One reference described in words: "Similar to [a brief visual description]" works better than abstract adjectives

Weak briefs produce technically competent images that solve the wrong problem. A complete brief produces images you can route for approval.

A product launch with a six-day deadline

Laura is a marketing manager at a 30-person direct-to-consumer supplement brand. She has a launch in six days and needs Instagram feed posts, a paid ad banner, and a landing page hero. Her freelancer is booked.

She wrote a visual brief: the product (matte charcoal canister, six inches tall, white label with green text) + placements (Instagram 1:1 and 4:5 feed, 16:9 banner) + audience (health-conscious 30s–40s who scroll fast) + tone (clean, confident, everyday — not clinical) + NOT (white-on-white studio photography) + required text (product name only) + one style reference described in plain terms (muted earthy background, natural afternoon light, single foreground object).

She ran the brief through ChatGPT Images in thinking mode. Twelve minutes later, she had six images. Three went into the campaign directly. One needed a crop. Two were wrong. She had what she needed by end of lunch.

The brief took longer to write than the generation took to run. That ratio is roughly correct.

When the images look good but don't go together

A content lead at a 90-person B2B SaaS company ran into a different problem. Her generated images were individually fine — good composition, correct text — but inconsistent as a set. Background palette shifted between cards. Lighting direction changed. The overall campaign looked like a folder of stock photos rather than a coordinated launch.

Her fix: she generated a style-setter image first. One image establishing only the visual environment — background texture, light direction, color temperature, no subject. She described this image in all subsequent prompts as "consistent with [this image]." The following six assets matched.

Coherent campaigns require a sequence, not just a thorough first brief. The style-setter step is not obvious until you've shipped a campaign without one.

What actually changed Monday morning

AI image generation that works across a campaign set is now a standard free tool. The teams that get useful output from it will spend their time on the brief, not the regeneration loop.