How to Create AI Images for Advertising Campaigns (Step-by-Step)
Wanderson Jackson
Updated August 2026. AI image generators can produce ad-ready visuals in minutes instead of days. This guide walks through the full workflow: choosing the right model, writing prompts that convert, formatting for each ad platform, and iterating on variants at scale. Whether you run Meta ads, Google Display, or TikTok Spark Ads, the process below covers what actually matters.
Before generating a single image, gather three things:
Your brand assets. Logo (PNG with transparent background), brand fonts, and a one-line value proposition. These feed into prompts and post-generation overlays.
Your ad platform specs. Each platform has different aspect ratio and file size requirements. Getting this wrong means your image gets cropped into nonsense.
A clear offer. AI can generate beautiful images. It cannot decide what your ad should say. Lead with the offer (discount, benefit, social proof) and build the visual around it.
Tools like the Avocado AI Workspace let you store brand assets and generate images from the same interface, so you skip the export-import dance between tools.
Step 1: Pick your AI image model
Not all image models work the same way for advertising. Some render text cleanly (critical for ad copy overlays). Others nail photorealism but butcher lettering. Here is what matters for ad images:
For ads with text overlays (headlines, CTAs, prices): Use a model with strong text rendering. GPT-Image 2 and Ideogram V3 both handle in-image text well. Seedream V5 Pro supports text in 14 languages, which matters for international campaigns.
For product-focused ads: Photorealism and clean detail matter more than text. Nano Banana 2, Recraft V4, and Krea 2 Medium all produce clean product shots. Recraft V4 also offers style control, which helps maintain visual consistency across a campaign.
For high-volume variant testing: Speed and cost matter. Nano Banana 2 Lite and Grok Imagine both cost 1 credit per image on Avocado AI and generate fast, which is ideal when you need 20-30 variants for A/B testing.
Avocado AI pricing runs from EUR 19.99 to EUR 249 per month, with credits allocated per plan. One-credit models cost roughly EUR 0.10 to EUR 0.13 per image on the Starter plan. That is cheaper than most stock photo subscriptions, and every image is unique.
Step 2: Know your ad format specs
Generating an ad image without knowing the destination platform is wasted effort. Here are the formats that matter:
Platform
Placement
Aspect ratio
Min resolution
File format
Meta (Facebook/Instagram)
Feed
1:1 or 4:5
1080x1080 or 1080x1350
JPG, PNG
Meta
Stories/Reels
9:16
1080x1920
JPG, PNG
Google Display
Banner
16:9, 4:3, 1:1
Varies by placement
JPG, PNG, GIF
Google Display
Leaderboard
6.73:1 (728x90)
728x90
JPG, PNG
TikTok
In-feed
9:16
1080x1920
JPG, PNG
TikTok
Spark Ads
9:16
1080x1920
JPG, PNG
LinkedIn
Feed
1.91:1
1200x628
JPG, PNG
Amazon
Sponsored Brands
Various
1000x1000+
JPG, PNG
Key rule: Generate at the largest resolution your model supports, then crop. Never generate at the exact minimum. A 1600x1600 source image can be cropped to 1080x1080 (Meta feed) or 1080x1350 (4:5) without quality loss. A 1080x1080 source cannot become a 9:16 Story without black bars.
Avocado AI supports aspect ratios including 1:1, 16:9, 9:16, 4:3, 3:4, 21:9, 3:2, and 2:3, which covers every major ad platform format.
Step 3: Write a high-converting ad image prompt
The prompt is where most people fail. A vague prompt like "product on a background" gives you generic stock-photo energy. A good ad prompt includes three elements:
1. Subject with context. Not just "a shoe" but "a running shoe on a wet city street at dawn, motion blur on the background." Context creates mood.
2. Lighting and atmosphere. This controls the emotional register. "Natural morning light, warm tones" reads differently than "dramatic side lighting, cool blue tones." Match the mood to your offer.
3. Composition direction. Tell the model where the subject should sit. "Centered, negative space on the left for text overlay" gives you room to add headlines in post-production.
Example prompt for a coffee brand ad:
"A steaming cup of artisan coffee on a rustic wooden table, morning sunlight streaming through a window, shallow depth of field, warm golden tones, negative space on the right side for text. Clean, editorial photography style."
Example prompt for a SaaS product ad:
"A clean, modern laptop on a minimalist desk showing a dashboard interface, soft diffused lighting, muted pastel tones, professional workspace setting, plenty of whitespace around the screen for overlay text."
What to avoid in ad prompts:
Do not ask for text in the image if you will overlay text in post. The model will render its own text that conflicts with yours.
Do not specify hex color codes. Let the model interpret natural-language descriptors for color and mood.
Do not reference real brands, logos, or recognizable products by name.
One image is never enough for an ad campaign. The workflow is:
Generate 5-10 variants from the same prompt with slight variations. Change the lighting direction, swap "morning" for "golden hour," adjust "centered" to "off-center." Small prompt changes create meaningfully different outputs.
Pick your top 2-3. Do not pick the one you like best. Pick the one that best serves the ad's purpose. A product ad needs the product clearly visible. A lifestyle ad needs emotional energy.
Post-process. Crop to platform specs. Add text overlays, logos, and CTAs in your design tool (Figma, Canva, or directly in the Avocado Workspace if you are composing there). Export at platform-required resolution.
Generate more if needed. The cost of an extra generation round is minimal. On Avocado's Starter plan (EUR 39/mo, 300 credits), 10 images at 1 credit each costs EUR 1.30 in credits. Compare that to a photographer's hourly rate.
Pro tip: Use Avocado Flows to build a repeatable pipeline. Set up your prompt template once, then run it with different subjects or moods for each campaign. This turns a 30-minute process into a 2-minute one for subsequent campaigns.
Step 5: Test variants at scale
The real advantage of AI-generated ad images is not one great image. It is 30 good images you can test simultaneously.
Testing framework:
Generate a matrix of variants. For a single product ad, create variations across two axes: visual style (lifestyle vs. product-close-up vs. UGC-style) and text overlay (benefit-led vs. social-proof-led vs. price-led). That gives you 9 variants from two prompt families.
Upload to your ad platform. Meta's Advantage+ campaigns, Google's responsive display ads, and TikTok's Spark Ads all accept multiple creative variants and optimize delivery toward the best performer.
Let the platform allocate spend. After 3-5 days and roughly 1,000 impressions per variant, you will see clear winners. Kill the bottom 50%, keep the top performers, generate new variants based on what worked.
Iterate weekly. Creative fatigue is real. Ad performance decays after 7-14 days on Meta and faster on TikTok. A weekly refresh cycle with 5-10 new AI-generated variants keeps performance stable.
Generating at the wrong aspect ratio. A square image cropped to 9:16 loses half the scene. Generate natively in the aspect ratio your platform needs.
Too much detail in the prompt. Overloaded prompts confuse models. "A woman, 28, wearing a blue jacket, holding a red bag, standing in a park, autumn leaves, golden hour, shallow DOF, bokeh, professional photography, 8K, HDR" is too much. Pick 3-4 key elements and let the model compose.
Ignoring negative space. Ad images need room for text overlays. If the subject fills the entire frame, there is nowhere to put the headline. Always direct the model to leave space.
Not testing enough variants. One image is not a campaign. Generate 10 minimum. The cost difference between 1 and 10 images is negligible; the performance difference is not.
Using the wrong model for the job. A model that excels at landscapes will not necessarily nail product photography. Match the model to the use case (see the table in Step 1).
FAQ
How much does it cost to generate AI ad images?
On Avocado AI, single images cost 1 to 2 credits depending on the model. The Starter plan at EUR 39/month includes 300 credits, which covers 150 to 300 images. That is enough for multiple ad campaigns with extensive variant testing.
Can I use AI-generated images in Facebook and Google ads?
Yes. Both Meta and Google accept AI-generated images in ad creatives. Meta's advertising policies require that images are not misleading and comply with their community standards. Google's policies are similar. Always check the latest platform policies, as they evolve.
What aspect ratio should I use for Instagram ad images?
For Instagram feed ads, use 1:1 (1080x1080) or 4:5 (1080x1350). For Instagram Stories and Reels ads, use 9:16 (1080x1920). The 4:5 format gives you more vertical screen real estate in the feed, which generally performs better for product ads.
How do I add text and logos to AI-generated images?
Generate the base image with AI, then add text, logos, and CTAs in a design tool like Figma, Canva, or Adobe Express. Some platforms like Avocado AI's Workspace let you generate images and compose layouts in the same environment. Do not rely on the AI model to render your brand text correctly.
How many ad image variants should I test?
Generate at least 10 variants per campaign. Upload all of them to your ad platform and let the algorithm optimize delivery. After 3-5 days with sufficient impressions, cut the bottom performers and generate fresh variants. A weekly refresh cycle prevents creative fatigue.
What is the best AI image model for product ads?
For clean product shots, Nano Banana 2 (1 credit on Avocado) and Recraft V4 (1 credit) both produce sharp, detail-rich output. For product ads with text overlays (prices, discount codes), GPT-Image 2 (2 credits) handles in-image text rendering more reliably than most alternatives.
If you want one workspace for generating, iterating, and managing AI ad images across campaigns, start with Avocado AI. Plans range from EUR 19.99 to EUR 249 per month with a full model catalog at every tier.
Written by Wanderson Jackson, founder of Avocado AI. I built Avocado to give creative teams one workspace for image, video, and audio generation instead of stitching together five different tools.
