GPT Image 2 for Ad Images

GPT Image 2 is one of two image models inside Polaris, and it is the one to reach for when your ad image needs legible text or tight, deliberate composition. It renders at 1K for 19 credits, 2K for 24 credits, or 4K for 29 credits, and when you attach reference images it automatically switches to its image-to-image variant so the output is built around your actual product rather than a lookalike.

What GPT Image 2 is best at for ecommerce ads

Two things separate it from most image models: text rendering and composition. If your static ad needs a headline baked into the image, a price flash, a badge like "NEW" or "BACK IN STOCK", or a label on the product itself that has to stay readable, GPT Image 2 is the safer pick. It also handles multi-element layouts well — product plus hand plus background plus copy — where looser models tend to drift.

That makes it a strong fit for:

• Static feed ads with on-image copy and offers
• Product-in-context shots where the label must stay accurate
• Before/after or comparison-style layouts that need clean structure
• Thumbnails and cover frames for your UGC video ads

Specs and credit costs

With reference images attached (passed as input_urls), the model runs in its image-to-image mode — useful when you need the generation anchored to your real packaging, colors, and shape.

When to pick it over other Polaris models

Polaris also offers Nano Banana Pro for images. The honest split: pick GPT Image 2 when text accuracy and composition control are the priority; consider the alternative when you want a different aesthetic character or are iterating on looser lifestyle scenes. Neither replaces testing — run variants of both and let the ad account decide. And if the concept really needs motion, a video model will usually outperform a static; see the full lineup on the models page.

How to use it

Polaris runs as an MCP connector inside Claude. Connect it once, then ask Claude in plain chat for the ad image and attach your product photo. When the settings question appears as tappable option cards, pick GPT Image 2 on the model card and choose your resolution. The live panel renders progress and the finished image in the conversation, typically in 20-180 seconds. The full setup is covered in this walkthrough.

Trade-offs to know

Higher resolution costs more credits, and 4K is rarely necessary for feed placements — 1K or 2K covers most ad slots. GPT Image 2 builds ads around your product, not a stock-avatar library, so results depend on the quality of the reference photo you attach. Your first batch of 12 is free with no card; after that, Pro is $49/mo for 2,000 credits — details on the pricing page.