TL;DR: AI image generators can produce display banner creatives across all IAB standard sizes in minutes instead of days. This guide walks through the full workflow from concept to export, covering size requirements, brand consistency, and batch production for Google Display Network, Meta Audience Network, and programmatic DSPs.
Display advertising still drives significant reach. Google Display Network alone reaches over 90% of internet users worldwide. Programmatic display ad spend exceeded $100 billion in 2025, and the format remains a core channel for retargeting, brand awareness, and mid-funnel campaigns.
The production problem is volume. A single campaign across Google Display Network and Meta Audience Network requires 5 to 8 banner sizes minimum. Add programmatic DSPs like DV360 or The Trade Desk, and the count climbs to 10 or more. Each size needs its own layout: a 300x250 medium rectangle reads differently from a 160x600 skyscraper. Text that works on a 728x90 leaderboard gets crammed on a 320x50 mobile banner.
Most teams handle this with one of two approaches: a designer manually resizes each creative (slow, expensive) or a template tool auto-generates variants from a single layout (fast but generic). AI image generation adds a third path: generate distinct visuals tailored to each format, maintaining brand consistency without templated repetition.
Standard Display Ad Sizes You Need to Cover
Before generating anything, know which sizes your networks require. The IAB (Interactive Advertising Bureau) defines standard display units that Google, Meta, and most DSPs accept:
Size (px)
Name
Where It Shows
Priority
300x250
Medium Rectangle
In-article, sidebar, mobile feeds
High
728x90
Leaderboard
Desktop header, footer, between content
High
160x600
Wide Skyscraper
Desktop sidebar
Medium
300x600
Half Page
Desktop sidebar, high-visibility
Medium
320x50
Mobile Leaderboard
Mobile apps, mobile web
High
970x250
Billboard
Desktop homepage, premium placements
Low
320x100
Large Mobile Banner
Mobile apps, higher fill rate
Medium
468x60
Banner
Legacy desktop placements
Low
Most campaigns cover the top 3 to 4 sizes (300x250, 728x90, 160x600, 320x50) and add 300x600 for high-value placements. The rest depend on your specific network and audience.
Each network also has file format and size limits:
Network
Max File Size
Formats
Animation
Google Display Network
150 KB (static), 150 KB (HTML5)
JPG, PNG, GIF, HTML5
Max 30s, max 5 loops
Meta Audience Network
Varies by placement
JPG, PNG, MP4
Video banners supported
DV360
150 KB (standard), 2.2 MB (rich media)
JPG, PNG, GIF, HTML5
Max 30s
Amazon DSP
150 KB
JPG, PNG, GIF
Max 15s
Keep these limits in mind during generation. A 1280x720 hero image is overkill for a 300x250 banner. You need purpose-built visuals, not just resized hero shots.
Step 1: Define Your Campaign Visual Direction
Before touching an AI tool, lock down three things:
Product or offer focus. What is the banner selling? A single product, a category, a seasonal promotion, or a brand message. Display banners have limited real estate. One focal point beats two or three.
Audience context. Where will these banners appear? Retargeting banners showing a product someone already viewed look different from prospecting banners introducing a brand for the first time. Retargeting can be product-forward with price and CTA. Prospecting needs a hook or visual story.
Brand constraints. Logo placement rules, brand colors, typography, and any required legal text (disclaimers, trademarks). These get baked into the overlay step later, but they influence the base image generation.
Write these down as a one-paragraph creative brief. It takes 5 minutes and saves you from regenerating 20 banners because the direction was vague.
Step 2: Generate Your Base Creative with AI
This is where AI image generation replaces the stock photo search and custom photoshoot.
Choosing the right model for banner ads. Not every AI image model handles small-format display banners well. You need clean composition, strong focal points, and the ability to work at non-standard aspect ratios.
Model
Credits per Image
Best For
GPT-Image 2
1 to 2
Text rendering on banners, photorealism, clean product shots
Recraft V4
1
Design-focused layouts, SVG output for scalable banners
Ideogram V3
2
Logos, posters, on-image text for CTA overlays
Nano Banana 2
1
Fast iteration on product-centric visuals
Seedream V5 Pro
2
Dense layouts with multilingual text
For display banners specifically, GPT-Image 2 and Recraft V4 are the strongest starting points. GPT-Image 2 handles text rendering well (important for banners with embedded CTAs), while Recraft V4 has native design-focused controls that work for ad layouts.
Writing the prompt for a banner. Banner images are not hero images. They need to work at small sizes, which means:
One clear subject, not a busy scene
High contrast between subject and background
Room for text overlays (headlines, CTAs) without covering the subject
Simple, bold composition that reads at 300x250 pixels
Example prompt for a product-focused retargeting banner:
"A single premium wireless headphone on a clean neutral surface, soft directional lighting from the left, generous negative space on the right side for text placement, shallow depth of field, product photography style, no text, no logos."
Example prompt for a brand awareness banner:
"A person in a modern home office wearing casual clothing, natural daylight from a window, candid moment adjusting their desk setup, warm tones, iPhone photography style, no text, no brand names, no logos."
Aspect ratios matter. Display banners are rectangular, often with extreme ratios. The standard AI image aspect ratios cover most needs:
4:3 works for 300x250 (close to 6:5, crop to fit)
21:9 or 16:9 works for 728x90 and 970x250 (wide leaderboards)
9:16 works for 160x600 (tall skyscrapers)
Generate at the closest available ratio and crop precisely in post. Avocado AI Workspace supports 1:1, 16:9, 9:16, 4:3, 3:4, 21:9, 3:2, and 2:3, which covers all display banner aspect ratios with minor post-crop adjustments.
Step 3: Resize and Adapt Across IAB Formats
A single base image will not work across all sizes. A 300x250 banner needs a tighter crop than a 728x90 leaderboard. A 160x600 skyscraper needs vertical composition, while a 300x600 half page can accommodate more breathing room.
The practical approach:
Start with your highest-priority size (usually 300x250 or 728x90).
Generate the base creative optimized for that format.
For other sizes, either:
Generate a new image with a modified prompt emphasizing vertical or horizontal composition, OR
Use the same base image and crop/resize, accepting some composition loss.
Generating distinct images per size is better but costs more credits. For a 4-size campaign at 1 to 2 credits per image, that is 4 to 8 credits total. On Avocado AI Starter at 300 credits per month, that is a small fraction of the monthly budget.
Batch approach in Avocado AI Workspace: Generate all size variants in one session. Use Compose Mode to set up a generation pipeline with your base prompt modified for each aspect ratio. Download all variants, then move to overlay.
Step 4: Add Copy and Brand Overlays
AI-generated images are the visual layer. Display ads also need:
Headline text (5 to 8 words max for small banners)
These are design overlays, not AI generation tasks. Use your design tool of choice (Figma, Canva, Adobe Express) to layer text and logos on top of the AI-generated base image.
Text hierarchy for small banners (300x250 and smaller):
Product image or visual hook (60 to 70% of the space)
Headline (large, bold, 1 line)
CTA button (contrasting color, clear action verb)
Logo (small, corner placement)
Text hierarchy for larger banners (728x90 and wider):
Visual on one side (left or right)
Headline + subheadline (middle)
CTA button (opposite side from visual)
Logo (integrated into the CTA area)
Avoid the common mistake of text overload. Display banners have under 2 seconds to register. If someone cannot read the headline and understand the offer at a glance, the banner fails regardless of visual quality.
Step 5: Export and Validate for Each Network
Before uploading to your ad network, verify:
File size. Google Display Network caps static images at 150 KB. Export as optimized JPG (quality 80 to 85) or compressed PNG. If your file exceeds 150 KB, reduce resolution or increase compression.
Format. Most networks accept JPG and PNG. GIF for simple animations. HTML5 for rich media (requires Google Web Designer or similar).
Dimensions. Exact pixel dimensions must match the ad unit. A 301x251 image will be rejected for a 300x250 slot.
Safe zones. Some networks have border requirements. Google Display Network requires a visible border on ads with white backgrounds (1-pixel gray border is common).
Quick validation checklist per banner:
Correct pixel dimensions?
Under file size limit?
Correct format (JPG/PNG/GIF)?
Text is legible at actual display size?
Logo is visible but not dominant?
CTA is clear and readable?
No misleading content or trademark violations?
Step 6: Test Multiple Variants for Creative Fatigue
Display ads fatigue fast. Average banner CTR across all formats is under 0.5%, and performance drops after 2 to 3 weeks of the same creative. The production speed advantage of AI generation is most valuable here: instead of running one set of banners for a month, you can rotate 3 to 4 visual variants every 2 weeks.
Practical variant strategy:
Generate 3 to 4 distinct base visuals for your campaign (different subjects, angles, or color temperatures).
Each variant covers all your IAB sizes.
Rotate variants every 2 weeks in your ad network.
Track CTR and conversion by variant. Kill the underperformers, generate replacements.
This is where Avocado AI Workspace earns its keep. Generating 12 to 16 banner images (4 variants times 4 sizes) at 1 to 2 credits each costs 12 to 32 credits. On the Starter plan at 300 credits per month, that leaves room for multiple campaign rotations within a single billing cycle.
For teams running display campaigns at scale across multiple products or categories, Flows lets you build reusable generation templates. Set up the base prompt once, parameterize the product and aspect ratio, and generate a full banner set with one workflow run.
What Actually Matters for Display Banner Production
Three things separate effective AI banner workflows from wasted credits:
1. Prompt specificity for small formats. Generic prompts produce images that look fine at 1280x720 and muddy at 300x250. Write prompts that emphasize a single focal subject with high contrast and generous negative space. Test at actual display size before committing to a full set.
2. Size-specific composition, not just cropping. A 728x90 leaderboard and a 160x600 skyscraper need different visual compositions. Generate separately for each format family (horizontal, square, vertical) rather than forcing one image into all sizes.
3. Variant rotation beats single-perfection. A "good enough" banner that gets rotated every 2 weeks outperforms a "perfect" banner that runs for 3 months. AI generation makes rotation cheap. Use that advantage.
FAQ
Can AI-generated images be used for Google Display Network ads?
Yes. Google Display Network accepts JPG, PNG, GIF, and HTML5 creatives. AI-generated images exported in these formats at the correct dimensions and file size (under 150 KB for static) are fully compliant. Ensure the image does not contain misleading content, trademarked elements, or text that violates Google Ads policies.
What AI model is best for creating display banner ads?
For display banners, GPT-Image 2 (text rendering, photorealism) and Recraft V4 (design-focused layouts) are the strongest options. GPT-Image 2 works well when your banner includes embedded text or product photography. Recraft V4 excels for design-forward layouts with clean geometric composition. Both are available on all Avocado AI tiers.
How many banner sizes do I need for a display campaign?
Most campaigns need 4 to 6 sizes: 300x250 (medium rectangle), 728x90 (leaderboard), 160x600 (wide skyscraper), 300x600 (half page), and 320x50 (mobile leaderboard). Google Display Network and Meta Audience Network cover these as their primary units. Add 970x250 for premium desktop placements if budget allows.
How do I keep brand consistency across AI-generated banners?
Start with a consistent prompt template that includes your brand's visual direction (mood, lighting style, color temperature) without hardcoding hex colors. Use the same base prompt modified per size. Add logos and brand typography in your design overlay step, not in the AI prompt. This keeps the visual layer consistent while letting the overlay layer carry brand identity.
Can I animate AI-generated banner images?
You can use AI-generated still images as frames in animated GIF or HTML5 banners. Tools like Google Web Designer or After Effects can add simple animation (fade transitions, element movement) to AI-generated base images. For video banner formats, Avocado AI video models like Seedance 2.0 can generate short animated clips that serve as banner content.
How often should I rotate display banner creatives?
Every 2 to 3 weeks. Display banner CTR drops significantly after the first 2 weeks of the same creative. AI generation makes it practical to maintain 3 to 4 variant sets and rotate them on a biweekly schedule without recurring designer costs.
If you want one workspace for generating, resizing, and iterating on display banner creatives across every IAB format, start with Avocado AI. Check out our pricing for details.
Written by Wanderson Jackson, founder of Avocado AI. Avocado AI is a creative workspace for generating images, video, audio, and marketing content with AI models.
