Luma AI has opened its image and video tools to people who work inside AI assistants. On 3 November 2026 the company announced an MCP connection that lets users of Claude and ChatGPT ask for images or video in an ordinary conversation and get the finished files back in the same chat. Claude receives it as a connector, while ChatGPT gets an official Luma plugin.
Luma frames the launch around marketing and agency teams, which it says come under pressure in the fourth quarter to turn one winning ad into a dozen variants. Its pitch is that the analysis of what worked and the production of what comes next can happen in one place, with no handoff between tools. It supports the idea with a figure about last year's Black Friday and Cyber Monday period, when advertisers who added fresh creative almost every week reportedly saw a 23% higher median seven-day return on ad spend.
Setup is short. In Claude, a user adds the connector under Customize, then Connectors, clicks Connect and signs in with Google, Apple or an email address, choosing which team's credits will be billed. Any Luma account works, and generations draw on the credits the user already has, so there is no extra subscription. On Claude Team or Enterprise plans, one admin adds the connector for the whole workspace and each person then signs in. Claude asks before using any connector tool, though Luma's can be set to always allowed. The setup keeps billing with the team that owns the credits, so a team that already pays for Luma sees the new use appear in the same account.
The connector works at three levels. Single model calls use Uni-1 for images and Ray 3.2 for video, for example a clip of five or ten seconds. Pre-set mini-app workflows cover Ad Variants, Background Swap, Camera Angles, Light Transfer and Face Swap. Ad Variants renders in five aspect ratios and five languages (English, German, Italian, Japanese and Korean) and pauses on a preview before the final renders. Background Swap takes one to eight images, Camera Angles up to eight, and Light Transfer works up to 4K. The third level is larger multi-step projects.
Cover artwork of Luma's guide to using its connector with Claude
Image: Luma (connector guide)
The assistant can also look through the team's Asset Library, use brand files as inputs and save results into a folder. It deletes something only when asked to. Everything it makes lands in the user's Luma team, where it can be picked up again in a Luma board. The example Luma gives is asking the assistant which of last week's ads did best, then having Luma restyle that ad with new scenes, camera angles and lighting without leaving the conversation. Because the results sit in Luma rather than only in the chat, a designer can take over a draft the assistant started and keep refining it by hand.
About Luma AI
Luma was first known as Luma Labs, a San Francisco company that built the Genie 3D model generator before it released its Dream Machine text-to-video model to the public on 12 June 2024.
Its team includes inventors of NeRF, a technique that turns photos of an object into a 3D representation, and its Ray3 video model was built into Adobe Firefly in September 2025. On 19 November 2025 Luma raised a $900 million Series C led by the Saudi company HUMAIN, with AMD Ventures, Andreessen Horowitz, Amplify Partners and Matrix Partners also taking part.
2021
Founded
2
Models indexed
Models on AIPopDB
Luma lists a separate MCP server for its events product, which is distinct from this creative connector. The new connector is aimed at production work with Luma's image and video models rather than event management.
Square example image from Luma's guide to its creative workflows
Image: Luma (connector guide)
Sources · 2