Skip to content

ai-image-gen

Target Audience: Marketers, content creators, and designers who need AI-generated visuals — social graphics, persona portraits, framework diagrams, ad creatives, or product mockups — without leaving the CLI.

  • Area: Marketing
  • Category: images
  • Level: advanced
  • Input: conversation
  • Output: document
  • Setup: api-key
  • Roles:

Generates images across Google Gemini, Google Imagen and OpenAI gpt-image, picking the model that fits the job and the key that is set.

How you ask

  • “Generate a product hero image using Gemini”
  • “Create a social media graphic with Nano Banana”
  • “Build a style library from these reference images”
  • “Iterate on this AI-generated image — make the background warmer”

Inputs

  • request — Visual direction, subject, and mood to explore for a first pass; requirements can stay loose at this stage.
  • request — The specific change to apply to the existing render — what to keep and what to adjust, one or two variables at a time.
  • request — Photorealistic subject or scene to render — product shot, portrait, or physical detail where realism matters most.
  • aspect-ratio — Output shape locked before rendering (e.g. 1:1, 16:9, 4:5) — Imagen is single-shot, so this is fixed upfront rather than iterated.
  • resolution — Resolution tier (1K, 2K, 4K) — applies only to OpenAI gpt-image size selection; Imagen and Gemini ignore it and render at the model's default size.
  • request — Subject to merge with the loaded or extracted style before rendering.
  • style-id — ID of a saved style from assets/style-library/STYLE-GUIDE.md to load and merge with the subject.
  • reference-image — Inspiration image to extract a reusable style from via extract_style, used when no saved style ID fits.

Outputs

  • Draft PNG showing an initial visual direction for the requested subject.
  • PNG rendered by merging the loaded or extracted style with the current subject, ready to promote to the style library once confirmed.
  • Photorealistic PNG rendered via an Imagen model at the locked aspect ratio.
  • Updated PNG produced by continuing the multi-turn session with the requested change applied.
  • prefer a permanent directory)