Make Ad Images with GPT Image 2 in Claude
Generate ad images with GPT Image 2 inside Claude via Polaris MCP: connect once, describe the shot, choose settings, get your ad still.
To make ad images with GPT Image 2, connect Polaris to Claude through the MCP connector, then ask Claude in chat for the image and attach a product photo or written brief. Claude shows the available settings as tappable option cards, and the finished image renders in a live panel right in the conversation.
What GPT Image 2 is for
GPT Image 2 is one of the image generation models available through Polaris, used for ad stills such as product photography, lifestyle-style scenes, and static concepts that can run as standalone ads or serve as a reference frame for a later video ad. It's available in the same chat-based flow as Nano Banana Pro, so picking between the two models is just one of the tappable options, not a different process.
Step-by-step: generating in Claude
- Connect Polaris once. Add the MCP connector at polarisugc.com/api/mcp inside Claude and authorize it. This only needs doing a single time.
- Give Claude the product. Attach a product photo, or describe the product and the ad concept in a short chat message.
- Ask in plain language. Describe the image you want in ordinary words, no special prompt formatting needed.
- Choose from option cards. Claude presents choices like model (GPT Image 2), aspect ratio, and other settings as tappable cards.
- Track the render. A live Polaris panel appears in the chat and shows progress, generally landing somewhere between 20 and 180 seconds before the finished image appears.
- Batch a few variants. Ask for parallel variations rather than one at a time, then download the set as a ZIP once you've picked winners.
The web app is available as a secondary option for anyone who wants a dashboard view, but the chat workflow inside Claude is the primary way to work with Polaris.
Pro tips
- Lock down what must stay accurate (product shape, label, colors) in your brief, and leave the rest (backdrop, lighting, camera angle) open for Claude to vary across a batch.
- Generate several images in parallel and compare rather than repeatedly refining a single result serially.
- Use GPT Image 2 stills as a starting frame if you plan to build a matching video ad afterward, so the still and the motion creative feel like one campaign.
- Run any on-image or accompanying ad copy through Copy Judge before publishing, particularly in regulated categories.
Honest caveats
GPT Image 2 generations through Polaris are built around your actual product, not a stock-avatar library, so the input photo or brief you provide matters. A good-looking still is still a hypothesis, not a guaranteed winner, so plan to test it against your current creative rather than assuming it will outperform on the first try. In regulated niches, keep claims compliant in the copy that accompanies the image, no disease, cure, or guaranteed-result language.
Read more on how the Claude MCP workflow is structured, look at example AI UGC ads, or review pricing to plan usage.