Product Updates··7 min read

On-Model Now Works in ChatGPT and Claude

On-Model as a ChatGPT plugin and Claude connector: put garments on a model, re-cast a product page from a link or build a template, in plain language.

By On-Model Team

A chat window on a dark background linked by blue lines to a grid of on-model fashion photos

Most of the work around a product image isn't the image. It's finding the right photos, picking the model, choosing the settings, starting the job and checking what came back. Now you can hand that to the assistant you already use.

On-Model is available in ChatGPT as an app, and in Claude as a connector. Connect it once, then ask in plain language: the assistant uploads your photos, picks a model, tells you what the job will cost, runs it in your On-Model account and shows you the results in the chat, with a link to the job in the app. It uses the same tools and the same quality checks as the app, and every job lands in your account like any other.

Results arrive as a gallery in the chat, one click from the job in the app.

How to add On-Model to ChatGPT and Claude

You need an On-Model account on a Pro or Enterprise plan. Connecting an assistant uses API access, which those plans include.

In ChatGPT

  1. Open Plugins in the sidebar and search for On-Model.
  2. Click Connect and sign in with your On-Model email and password.
  3. In a chat, ask for On-Model by name, or pick it from the tools menu.
The same results gallery in ChatGPT.

In Claude

  1. Open Customize → Connectors in the sidebar.
  2. Click Add → Add custom connector, name it On-Model and paste https://mcp.on-model.com/mcp as the server URL.
  3. Click Continue, then Connect, and sign in with your On-Model email and password.
  4. In a chat, make sure On-Model is turned on in the tools menu.

Using Claude Code, Cursor or another tool that speaks MCP? The setup guide has the one-line commands.

Signed up with Google? Your account has no password yet. Set one with Forgot password on the sign-in page (Google sign-in keeps working), or connect with an API token from your profile.

Three things to ask it

1. Put an outfit on a model

Start from flat-lay photos of one outfit and say what you want. The assistant states the cost before it runs anything, and waits for your go-ahead on larger jobs.

Inputs
Shirt
Trousers
Mules
Result
Instruction: Here are flat-lay photos of an outfit. Put it on a female model, full body, on a light grey studio background, at 2K. Tell me the cost first.

The assistant chose a model from the library, wrote the shot and ran Flat-to-Model. When the automatic quality check flags something, it tells you and can fix it: ask it to try again with a styling note, or to retouch the flagged image.

2. Re-cast a whole product page from a link

Paste a link to a product page and ask for a different model. The assistant reads the page, collects the gallery photos, uploads them and runs Model Swap on all of them as one consistent set. Ask it to keep the clothes and the background identical, and only the model changes.

Before: the product page
After: same page, new model

"Here's our product page: on-model.com/demo/shop/logo-tee. Swap the model on every photo with Yumi from the library, and keep the clothes and the background exactly as they are."

It works with your own shop or a marketplace listing, as long as the assistant can open the page. If a site blocks it, paste the image links or attach the photos instead.

3. Build a template, then refine it

Templates are how On-Model keeps a look consistent across a collection. You can now build one in conversation. Share a few photos of the look you want and describe the set. The assistant writes each shot, tests it on your garments and shows you the results.

The reference photos

"Build a summer lookbook template in the style of these three photos: a full-body shot in the courtyard, a walking shot by the canal and a closer shot by the sea. Test it on the shirt and trousers."

First try

Then say what to change, the way you would to a photographer:

"Too golden, make it natural afternoon daylight so the colors stay true. A calmer expression, lips closed. And crop the third shot tighter on the collar."

After one round of feedback

When it looks right, say "save it as Summer Lookbook". The template appears under Templates in the app, ready for every product in the collection, from the chat with a bulk run or from the app. Later you can ask for one shot to change, re-test only that one and update the template in place.

Can Claude generate images?

Not on its own. Claude reads and describes images, but it does not create them. With the On-Model connector it can. Ask for an image from a prompt, an edit of a photo you share, a combination of several photos, or one of your saved models placed in a new scene, and Claude runs Image Playground and shows the result in the chat. Keep going turn by turn: "warmer light", "now a close-up of the bag", "same scene, different model".

"Create a 4:5 image of a linen shirt on a hanger by an open window, soft morning light, at 2K."

For product work, the same connector gives Claude everything above: putting garments on a model, packshots, colorways, model swaps and templates, all with the automatic quality check. ChatGPT can generate images on its own, but On-Model adds what a catalog needs: your garments reproduced as they are, consistent models across a set, and every result saved in your account.

Good to know

  • You confirm the cost. The assistant tells you what a job costs before starting anything large. Bulk runs always start with a free preview of how many images you'll get.
  • Everything stays in your account. Jobs, results and templates made from the chat appear in the app, so your team can review, retouch and download them there.
  • Images you already have work too. Name a product or a tag and the assistant finds the photos in your account, no new upload needed.
  • Your conversation stays with your assistant. On-Model only receives what the assistant sends to run a job: the images, the settings and the instructions. Never the chat itself. You can disconnect at any time from your assistant, or by revoking the token in your On-Model profile.

Common questions

Is On-Model a ChatGPT plugin? Yes. On-Model is listed in ChatGPT's Plugins directory. Search for On-Model, click Connect and sign in with your On-Model account.

Is there an On-Model connector for Claude? Yes. Add it as a custom connector in Customize → Connectors with the server URL https://mcp.on-model.com/mcp. It works in Claude on the web, desktop and mobile, and in Claude Code.

What is the On-Model MCP server? MCP (Model Context Protocol) is the open standard AI assistants use to connect to other software. The On-Model MCP server at https://mcp.on-model.com/mcp gives any MCP client, including ChatGPT, Claude and Cursor, the same tools as the On-Model app and the API.

How do I add an MCP server to Claude? Open Customize → Connectors, click Add → Add custom connector, give it a name and paste the server URL. For On-Model, the URL is https://mcp.on-model.com/mcp. Claude then asks you to sign in.

Which plans can connect? Pro and Enterprise. Connecting an assistant uses API access, which those plans include. Jobs use credits exactly as they do in the app.

Does On-Model see my conversation? No. On-Model only receives what the assistant sends to run a job: images, settings and instructions. The chat itself stays with your assistant.

Try it

Connect On-Model in ChatGPT or Claude and start with something you'd normally do by hand this week. If you'd rather call On-Model from your own code or agent, the API integration guide and how AI agents run a photo pipeline show the same tools without a chat. If you're not on Pro yet, compare the plans.

mcpchatgptclaudeai-agentsflat-to-modelmodel-swaptemplatesproduct-updates