Image generation

PulseAd’s image creative splits into two jobs: generate the image, then vary it into the shapes a campaign needs. Two tools cover them.

Generate — banana-bunch

A web app that turns a prompt plus optional product/reference images into many images at once. Choose between the default Nano Banana Pro model (Google Gemini Image API) and GPT-5.5 Image Generation (OpenAI Responses API).

• Batch generation — up to 20 images per job, produced by a worker pool (about 5 at a time by default via MAX_PARALLEL), so you pick from a spread rather than re-rolling one at a time.
• Image inputs — up to 10 product images and up to 10 style/reference images.
• Prompt extraction — upload an existing image and it writes the generation prompt that would reproduce that look, to reuse or tweak.
• Prompt translation — translate prompts between English and Korean.
• History and delivery — review generation history, download individual images or a ZIP, and download a 2× upscaled PNG.

Text-only GPT-5.5 requests use image-generation mode; when product or reference images are included, the app uses image-edit mode. The default Nano Banana Pro model remains available for the Gemini path.

The model to carry: it’s a spread generator, not a one-shot tool — you ask for many and curate, you don’t expect the first image to be final.

Vary — ads-variation

Once you have an image, a campaign needs it in several placements. ads-variation reflows a source design into other sizes — banner layout reflow and custom-size output — rather than regenerating from scratch, so the look stays consistent across placements.

Scope note

These are creative-generation tools. They produce assets; they don’t push them to Amazon — that’s the campaign/capability surface (see Capabilities). Examples use placeholder products only (Brand A, Product A).