Marketing Image Producer

What it does

Turns a marketing-image brief into a production plan without installing anything: which tool, which prompt, which dimensions, how to optimize. Output arrives in four blocks.

1. Tool selection (decision tree). From goal + constraints it picks one of five approaches — AI Generation, AI Editing, Design Tools, Screenshot+Overlay, Stock — and applies an up-to-date model matrix: for in-image text/headlines use Ideogram 3.0 (~90-95% text accuracy, clear category leader); for brand/product consistency use Flux (multi-image reference) or Recraft V3 (vector); for photorealism Flux Pro 1.1 / Imagen and Gemini Nano Banana Pro (multimodal reasoning); for in-place editing Gemini native / Flux Kontext. Hard rule: never generate product UI with AI — models hallucinate interfaces; use real screenshots + device mockups.

2. Prompt engineering. For each visual it writes a ready prompt using Subject + Setting + Style + Lighting + Composition + Technical, always specifies aspect ratio and a style direction ("photorealistic / flat illustration / 3D render"), and defers complex text to overlays.

3. Platform specs. 1200x630 for blog/OG, a per-platform table for social (X 1200x675, LinkedIn 1200x627, IG 1:1/4:5/9:16), banners for LinkedIn 1584x396 · X 1500x500 · Product Hunt 1270x760 · App Store/Play, and a Canva Magic Resize flow for many variants from one design.

4. Optimization + distribution. WebP/AVIF with JPEG/PNG fallback (<picture>), resize to display size, target quality 75-85%, loading="lazy", explicit width/height (prevents CLS), CDN auto-format, alt text, plus the OG meta-tag set and dynamic OG (Vercel OG / Satori).

When to use it

When a blog post needs a hero; when producing social/launch graphics; when preparing profile/directory banners (Product Hunt, G2, LinkedIn); when setting up OG/social preview images; or when you already have an image and need optimization/format conversion for web performance. For paid ad creative it defers to ad-creative, for video to video.

Method / frameworks

Model-selection decision tree (text → consistency → editing → vector → quality → cost), a 6-part prompt formula, platform dimension matrices, a Core Web Vitals-oriented optimization checklist (don't lazy-load the hero for LCP + fetchpriority; explicit dimensions for CLS), a format comparison table (WebP default ~97% support; AVIF smaller but INP risk on low-end), and the OG/Twitter meta contract. Adapted from coreyhaines31/marketingskills (MIT).