How to make AI influencer photos inside ChatGPT or Claude

To make AI influencer photos inside ChatGPT or Claude, save your AI model as a persona on PixelPig, add the PixelPig connector to the assistant, and ask for photos of that persona by name. The assistant does not invent the face. It passes your prompt to PixelPig, which renders the persona you saved, so the same face comes back in every chat. If you have no persona yet, the step-by-step guide to creating an AI influencer covers that part. For scene and skin detail in your prompts, see how to make AI photos look real.

Proof

We connected ChatGPT to the PixelPig server on production on 21 September 2026, and ran the Claude sign-in on production in the same release. For this page we then read the tool list and every limit from the server source, and the connect steps from the code behind the /mcp page. Each row below is what the server does today. The connector is part of the Pro and Scale plans.

Three details from the code matter in practice:

• The cost gate is mandatory. Every tool that spends tokens first answers with an estimate and your balance. It only generates when the assistant calls it again with your approval. For video and edits, the approved number travels with the second call, and the server refuses the job if the real cost is higher than the quote.
• A retried video does not start twice. ChatGPT can retry a call that runs long. The same video prompt confirmed twice within 3 minutes returns the first clip instead of starting a second one, and photo calls stay under ChatGPT's time limit, so they finish before ChatGPT's 60 second cutoff.
• The default image model is Nano Banana 2. Name another model in your message to switch. The list is Nano Banana Pro, Nano Banana 2, GPT Image 2, Seedream 4.5, Seedream 5 Lite and Grok Imagine.

Step 1: save the persona on pixelpig.app

The connector reads personas; it does not make them. Open Personas, tap Create Persona, and either pick a ready-made AI model or upload one clear, front-facing face photo of a consenting adult. If you build from real photos, our guide on how many photos you need to train a model of a person explains what a good face photo looks like.

Give the persona a short, unique name. In a chat you will call her by that name, and a name like "Camila" is easier for the assistant to match than "persona 3 final".

Step 2: add the connector to ChatGPT

ChatGPT calls custom MCP servers plugins, and adding one needs Developer mode. OpenAI documents the flow in its guide to connecting a plugin.

1. In ChatGPT on the web, open Settings, then Security and login, and turn on Developer mode. It comes with Plus, Pro, Business, Enterprise and Edu.
2. On the Plugins page, click the plus button, name it PixelPig, paste https://mcp.pixelpig.app/mcp, choose OAuth, and click Create.
3. ChatGPT opens PixelPig. Sign in and click Authorize.
4. In a new chat, click the plus button, open More, and pick PixelPig.

We ran this flow end to end on production on 21 September 2026. ChatGPT asks you to confirm each tool that spends tokens, because the server marks those tools as write actions. The read-only lookups (personas, balance, models, status) run without that prompt.

Step 3: or add it to Claude

On claude.ai the same server is a custom connector. Anthropic's custom connector guide lists them on Free, Pro, Max, Ultra, Team and Enterprise plans, with Free limited to one custom connector. On Team and Enterprise, an owner adds the connector first, then each member connects.

1. Open Customize, then Connectors, then Add custom connector. The "Add to Claude" button on our /mcp page opens this form with the name and URL filled in.
2. Click Continue, then Connect. Claude opens PixelPig in a new tab.
3. Sign in and click Authorize.

We ran the Claude sign-in on production. Cursor and Claude Code use the same URL and the same browser sign-in; the steps for both sit on the /mcp page. We have not yet run the full Cursor flow ourselves.

Step 4: ask for photos the way you would brief a photographer

Once connected, a first message like this works:

List my personas, then make three photos of Camila in a sunlit cafe, oversized cream sweater, holding a latte, 3:4.

The assistant calls list_personas, then calls generate_persona_photos once without confirming, which returns the token cost and your balance. Say yes and it calls the tool again; the photos come back in the chat and in your Gallery. Photos come back in the same answer unless one is still rendering. Video takes 1 to 5 minutes, and the assistant checks on it about every 45 seconds.

Describe the scene, not the face. The persona carries the face, so words about eye colour or face shape add nothing she does not already have. Spend the words on setting, outfit, pose, framing and light.

From there the chat does what a studio tab does. Ask for an edit of a photo you like ("same shot, evening light"), or ask to animate it into a short vertical clip for reels. For dance and trend clips, the AI dance video guide explains the source clip rules first.

Chat or studio: which to use

The two work well together. Plan a week of posts in a chat, generate the safe-for-work batch there, and open the AI image generator for the shots that need hands-on tweaks. Plans and token costs are on the pricing page.

Common mistakes

• Connecting before you have a persona. Photo generation needs a persona, so the assistant has nothing to work with. Make the persona first.
• Writing face details in every prompt. It fights the persona instead of helping it.
• Approving the cost without reading it. A 15 second clip at 1080p costs far more than a photo. The quote is there so you see that before you say yes.
• Expecting uncensored output in a chat. The connector is safe for work only. Use the studio.
• Pasting a page link instead of an image link. edit_image and face_swap need a direct image URL. Use the full-resolution link the assistant returned for an earlier photo.
• Staying connected after a downgrade. The connector stops generating when you leave Pro or Scale, or cancel. Your personas and Gallery stay.

Bottom line

The connector does not make ChatGPT or Claude better at faces. It hands the face to a persona you already saved, so the assistant only has to describe the scene. Save the persona, connect once, check every cost quote, and keep uncensored work in the studio. The AI persona page explains how personas hold a face across the image and video tools.