Integration · Create and promote

Replicate and Meteor AI agents

Generate and edit images and video from a Meteor Met: Nano Banana, GPT Image, FLUX, Veo 3.1, Seedance and more, with no keys and no accounts to configure.

Book a demo See every integration

How does Replicate connect to Meteor?

Meteor uses Replicate so your Met can generate an image from a description, edit an existing image and generate video, across some twenty models you pick by name: Nano Banana and GPT Image 2 for general imagery, FLUX Kontext Pro to edit from a reference, Ideogram v3 when the text inside the piece has to come out right and Recraft v3 for brand work; Veo 3.1, Seedance, Kling, Hailuo 02, Wan 2.1, LTX and Grok Imagine for video. Meteor supplies the provider key, so there is no account to create and no credential to configure. Usage is drawn from the Energy balance of your workspace.

What it automates

What a Meteor Met does with Replicate

The missing asset for the post

Your Met writes the copy for the post and generates the image that goes with it, then publishes it to Instagram or the blog without a detour through another tool.

Adjust instead of starting over

When the image is almost right, you edit it with instructions on that same asset instead of generating ten variants and picking one. That is what FLUX Kontext Pro is for.

Short video, sound included

A product photo becomes a short video by using it as the first frame, and on the models with native audio the sound comes out of that same request, not a separate edit.

Choose the model on purpose

Recraft v3 for a logo, Ideogram v3 for a piece with text and LTX to try something cheaply are not interchangeable. The listing says what exists, so the choice is not frozen into a prompt.

4 tools

Everything your Met can ask Replicate for

Each one is a real action in the connector. You can switch off the ones you would rather it never used.

Tool	What it does
`replicate_list_models`	Lists the image and video models available and flags which accept a reference image.
`replicate_generate_image`	Generates an image from a description, with the model you pick or the default one.
`replicate_edit_image`	Edits or transforms an existing image following the instructions you give.
`replicate_generate_video`	Generates video from a description and, if you want, from a first frame image.

The limits

What the Replicate integration does NOT do

We would rather say it here than have you find out during rollout.

It sends a single reference image per generation. Several models accept more: Nano Banana takes up to fourteen, and the omni-reference video models a good deal more. The chat tool still passes one, and it passes no reference videos or audio either.
It does not produce voiceover or music as a separate track. The only audio is what some video models generate along with the picture. Voiceover comes from other tools on the platform.
It does not expose the fine-grained model settings. You choose model, aspect ratio and duration. Resolution and the native audio switch stay on each model's own default.
It does not edit video that was already generated. It generates from a description or from a first frame, and that is where it stops.
It does not guarantee usage rights over what it generates. The result depends on the model you choose and its terms.
It does not let you bring your own provider key. The key belongs to the platform and usage is billed as workspace Energy.

Before you start

What you need to connect Replicate

1

Nothing on your side

The provider key belongs to Meteor and lives on the server, never in the database and never in the interface. All you need is Energy available in the workspace.

Frequently asked questions

Questions about Replicate and Meteor

No. The key belongs to the platform and lives on the server. What you use is drawn from your workspace Energy, the same as the rest of the AI.

For images: Nano Banana in its three versions, GPT Image 2, the FLUX family (1.1 Pro, Dev, Schnell, and Kontext Pro for editing from a reference), Ideogram v3 when the text inside the image has to come out right, and Recraft v3 for logos and brand assets. For video: Veo 3.1 and Veo 3 Fast, ByteDance's Seedance, Kling v2.1 Master for cinematic work, Hailuo 02, Wan 2.1, which always starts from an image, LTX as the cheapest option, and Grok Imagine. One tool lists them with their identifier, and checking it costs nothing.

Not as a separate track. What it does have is video with native audio: Veo 3.1, Veo 3 Fast and the newer Seedance models produce it along with the picture, in sync. That audio comes out on each model's own setting, because the chat tool cannot switch it on or off, and voiceover or music are made with other tools on the platform.

Yes, with the models that accept a reference image. You pass the image and the instructions for the change and it returns the edited version. One image per request.

On most models, yes. On the ones built to hold a character, that same image is taken as an identity reference rather than a starting frame: the video resembles the photo without beginning exactly on it.

It depends on the model, which is why you choose one. An image on FLUX Schnell costs a fraction of one on Nano Banana Pro, and a second of LTX a fraction of a second of Veo 3.1. It is all drawn from workspace Energy and recorded per generation.

Step by step