AI Image Generation for Ad Creative: Test 20 Variations in 2 Hours

Six months ago, a local HVAC company we work with was spending $850/month on a freelance designer to create variations of their Google and Meta ads. They'd request three versions, get them in a week, run them for two weeks, then repeat. Slow feedback loops meant they were always 2-3 weeks behind on learning what worked. We switched them to Midjourney and Ideogram for generating ad creative variations. Now they generate 15-20 variations in a single afternoon, test them within 24 hours, and pause underperformers after 48 hours instead of two weeks. Their cost per lead dropped 34% because they're testing creative much faster and scaling winners sooner. The trade: they spend 30 minutes writing detailed prompts instead of waiting for design feedback. That's a deal.

Set Up Your AI Image Generation Workflow

Start with a tool choice. Midjourney ($20/month) excels at photorealistic images and lifestyle scenes—great for service businesses. Ideogram ($10-30/month) is faster for text overlays and simpler composition. DALL-E 3 ($15/month or per-usage) works well integrated into ChatGPT. We recommend Midjourney for 80% of SMB ad work because the image quality is highest and it handles composition detail better. Create a Midjourney server (Discord-based) and invite your team. Set a budget cap to avoid runaway spending. Then build a prompt template that you'll use repeatedly.

Your prompt template should include: (1) what the image shows, (2) the mood/style, (3) who's in it, (4) what they're doing, (5) lighting and color palette. Example for a plumbing service: 'Photorealistic image of a 40-year-old male plumber smiling while installing a new water heater in a modern kitchen, warm natural light from window, kitchen is bright and clean, professional attire, focused expression, 4K quality.' Run this through Midjourney four times—you'll get four different interpretations. Each takes 60 seconds to generate. You've now created four ad variations in three minutes for $0.04 in generation costs.

• Sign up for Midjourney (or Ideogram if you prefer speed). Budget $20-40/month for testing.
• Write 4-5 core prompt templates for your business (one for 'hero image,' one for 'customer testimonial style,' one for 'before/after,' etc.)
• Generate 4 variations of each template (batch prompt with different seeds or slight prompt tweaks)
• Export all images to a shared folder, label by prompt version, and sort into Figma or Google Sheets for easy reference

Create Ad-Ready Variations That Test Different Angles

Don't generate 20 images of the same thing. Generate images that test different angles: emotional vs. practical, lifestyle vs. product-focused, faces vs. no faces, before/after vs. benefit-focused. A dental practice should test: (1) a smiling patient showing off teeth, (2) a dentist performing a procedure (trust), (3) a before/after comparison, (4) a family laughing together (emotional appeal). These are strategically different. When you run them as split tests on Google Ads or Meta, you'll see 40-60% variance in CTR and conversion rate depending on angle. The AI generation cost is the same ($0.20 for all four), but the learning is huge.

Use Figma or Canva Pro to add copy overlay to your AI images. This takes 10 minutes per image once you're comfortable. Most AI image tools don't handle text well, so this manual step is worth it. Add your headline, CTA, and any legal disclaimers. Export as 1200x628 (LinkedIn), 1080x1080 (Instagram), and 1200x800 (Google Display) without resizing the image—just the canvas. You're now running platform-optimized variations from the same core creative.

• Generate images representing 4-6 different emotional/benefit angles (not just style variations)
• Add copy overlay in Figma or Canva (headline + CTA). Takes 8-10 min per image.
• Export to three formats: square (Instagram), landscape (Google/LinkedIn), vertical (TikTok/Reels) if needed
• Label files with version name, angle tested, and date generated. Makes analysis easier later.

Test, Measure, and Scale Systematically

Run your 12-15 variations as a single campaign with different ad sets for two weeks. Don't rotate evenly—let Meta or Google optimize. After 14 days, pull performance data: CTR, CPC, conversion rate if you can track it. You'll usually see 2-3 clear winners (30-50% better CTR than the baseline). Pause everything except the winners and the second-best performer. Now generate new variations of the winning angle. If 'before/after' image with emotional headline outperformed others by 40%, generate four new before/after variations with slightly different prompts and copy. This is called 'winner validation.' Most SMBs skip this step and kill their best-performing creative too soon. One ecommerce client we work with tested 18 variations, found a clear winner (40% higher CTR), then generated five more variations of that angle. The second round of variations stayed within 8-12% of the original winner's performance, which meant they validated the angle and had fresh creative to rotate. They maintained top performance for four months before fatigue set in.

The speed of testing is your competitive advantage. If you can generate 20 variations and test them in 48 hours while a competitor is waiting two weeks for designer feedback, you learn faster. Learning faster is how you lower cost per lead.

Budget: allocate $300-500/month for ad spend split testing (not generation cost—that's minimal). Run each variation with a $20-30 daily budget for 14 days. Collect data, scale winners to $50-80/day, kill losers. Repeat every two weeks. Within two months, you'll have tested 40-50 variations and know exactly which images and angles drive the best results for your business. A service business that runs this system typically sees 25-35% improvement in CTR and 15-25% improvement in cost per conversion by month two, purely from creative optimization.

• Run 3 ad sets with 4-5 variations each (12-15 total). Daily budget $20-30 each, two-week test window.
• Pull CTR, CPC, and conversion metrics on day 7 and day 14. Note clear winners (top 30%).
• Pause underperformers. Scale winners to 2x budget. Generate new variations of winning angles.
• Repeat every 14 days. Document results in a simple spreadsheet (angle name, CTR, CPC, winner Y/N).

Legal and Disclosure: Disclose AI Where Required

FTC guidance is still evolving, but the safe approach: if your ad features a person or real scenario and it's AI-generated, disclose it. Add small text: 'Image generated by AI' or 'Illustration.' If it's clearly stylized or abstract, most platforms don't require disclosure yet. Check your platform's policies (Meta, Google, TikTok all have evolving rules). For B2B and service industries, we recommend disclosing because it builds trust—customers see 'AI-generated image' and think 'this company is modern and efficient,' not 'this is fake.' Don't hide it. One law firm tested ad creative with and without 'AI-assisted imagery' disclosure. The disclosed version had 8% higher CTR, suggesting transparency actually helps. Include the disclosure in the main copy or as a small hashtag (#AIGenerated).

← Back to all articles