How to Run an AI Fashion Photoshoot Without Leaving ChatGPT or Claude

Ask ChatGPT or Claude for an AI solution for fashion photoshoots and you usually get a list of tools to go sign up for. Connect the Uwear MCP server and the assistant stops recommending and starts producing: it reads your garments, picks a model, plans the looks, and comes back with one costed brief. You approve it, the shoot runs, and the finished product photos land in your Uwear workspace. You never leave the chat.

This guide walks the real flow on real screens, then shows the install for each client: ChatGPT, Claude, Claude Code, and Codex. All four connect to the same remote server with the same Uwear login, so start with the one your team already has open.

What a Fashion Photoshoot in Chat Looks Like

The whole run is three moves: ask, confirm the context, approve the cost. These screenshots come from one real session with the Uwear MCP connected.

Step 1: Ask for the shoot

Start with one message. No brief template, no upload form, no new tab. The run in these screenshots started with a single line asking for an urban photoshoot.

Step 2: Confirm the context in chat

The assistant calls the Uwear tools and opens your actual workspace inside the conversation: your clothing, your models, your credit balance. Drag the pieces you want onto the Shoot Board and confirm. On a local agent like Claude Code or Codex, this step can come straight from a product folder or a line sheet instead.

Step 3: Approve the costed brief

The assistant does the production thinking and comes back with an itemized brief: the looks it paired, the model, the framing, the art direction per look, and the exact credit cost of the run. Adjust any line, or confirm and let it run. Nothing spends a credit before you approve.

After you confirm, generation runs on Uwear. Results land in your workspace, in the same library Studio uses, so you can keep iterating in chat, review everything in Studio, or pull the files through the API.

Run It From ChatGPT

In ChatGPT, Uwear MCP installs as an app in Settings. Developer mode is required today.

Run It From Claude

In Claude, Uwear MCP installs as a custom connector, and it works on both Claude web and Claude desktop.

Run It From Claude Code

Claude Code runs on your machine, which changes what the photoshoot can touch: the agent reads garments straight from your product folders and writes finished images back to disk, exactly where your pipeline expects them.

claude mcp add --transport http uwear https://api.uwear.ai/mcp

Run It From Codex

Codex connects with two lines of configuration and works straight from your folders, CSVs, and line sheets.

# ~/.codex/config.toml
[mcp_servers.uwear]
url = "https://api.uwear.ai/mcp"

Using a different agent? Any client that supports remote MCP servers connects with the same URL and OAuth login, and the Uwear MCP page has a copyable install prompt the agent can follow on its own. Agents that prefer flat files can read llms.txt or UWEAR.md, both kept in sync with the product.

What to Ask For Once Connected

Ask for work, not for features. Each of these is one message in the chat where you installed Uwear, and each one stops at a costed brief before credits are spent.

Turn On QA and an AI Reviewer Checks Every Result

QA rides in the same conversation, and it is a switch you flip, not a gate you wait behind. Turn it on, and the AI reviewer checks each result against your model, your prompt, and every garment in the look. What passes ships. What fails retries with the reason attached, like a jacket rejected for showing five buttons where the reference has four, then retried and approved.

Retries are generations, so they cost credits, and the brief prices all of it up front. The 76-credit run above already included QA with one retry per look, so the number you approve is the number you spend. The QA and retries page shows how the AI reviewer works in detail.

Start on the MCP Page

Everything in this post starts at the Uwear MCP page: install tabs for every client, the copyable agent install prompt, and the FAQ on approvals, credits, and where results land. If you would rather build against HTTP than chat, the same production runs over the REST API. The agent skills page packages the flat-file references for agents that read documentation instead of tools, and Agent Mode runs the same conversation inside the product, no external assistant needed.