The right way to make a marketing image depends on the job: use AI generation (Gemini/Nano Banana, Flux, or Ideogram) for original heroes and social scenes, design tools (Canva, Figma) for brand-consistent templates, and real screenshots for product UI — never AI, which hallucinates interfaces. For AI prompts, follow Subject + Setting + Style + Lighting + Composition + Technical, keep them 40-80 words, always specify an aspect ratio, and use Ideogram when text must appear in the image. Blog heroes and OG images are 1200x630; then optimize every image by serving WebP with a JPEG fallback, resizing to display size, compressing to 75-85% quality, lazy-loading below the fold, and setting explicit width/height to prevent layout shift.
What's inside
The playbook covers
Key data
The recommendations are grounded in benchmarks, not opinion:
See it in action
I need a hero image for a blog post about email deliverability. Make it visually striking.
A clean workspace with a laptop displaying a colorful analytics dashboard, minimalist desk with a coffee cup and notebook, bright natural window lighting from the right, shallow depth of field, commercial photography style, 1200x630, high resolution
cwebp -q 80 hero.png -o hero.webp # target <200KB, serve WebP with JPEG fallback
FAQ
Can I use AI to generate an image of my product's dashboard?
No — AI models hallucinate UI, so the result won't match your real product. Capture real screenshots at 2x resolution instead (browser DevTools, Shottr, CleanShot X, or the screencapture CLI), frame them in device mockups, and add callout arrows or feature labels for context. Using AI for product UI is the single most common image mistake.
What size should a blog hero or OG image be?
1200x630 (a 1.91:1 ratio) works for both the blog hero and the OG/social preview image, which is the universal web standard. Use 1920x1080 (16:9) for a full-width website header. Add the og:image, og:image:width, og:image:height, and twitter:card meta tags so shared links render a preview.
Which AI model is best for images with text or headlines?
Ideogram 3.0 has the best text rendering of any model, followed by Gemini and GPT Image. Even so, keep text requests to 3-5 words for reliability — every model struggles with paragraphs. For anything longer, generate the base image without text and add the copy as an overlay in post.
Related skills
Build a tool for this workflow?
Marketers run this skill mid-task — the exact moment your tool matters. Partner tiers put your tool inside the skill their agent runs.
See partner tiers