Workflow · Setup Guide

Run AI image and video models
from one chat.

Connect Claude to fal and to your project folders. From then on you describe what you want in plain words, Claude picks the model, runs it on fal and saves the result straight into your folder. No nodes, no extra web interface.

Jump to the steps ↓

How it works

fal hosts basically every available image, video and language model, plus its own like FLUX 3 Max, and you only pay per generation. The fal MCP connects that catalog to Claude. Linking a local folder on top means Claude can read your inputs and write every output next to them, with file names you choose.

Workflow · 5 Steps

Step by step.

01Create your fal account

fal.ai

Sign up at fal.ai, add a payment method and set a spending limit under Billing. fal is pay as you go, there's no subscription. You only pay for what you generate.

Note: Set the limit before you start experimenting.

02Add fal as a connector in Claude

Claude Settings

In Claude go to Settings → Connectors → Add custom connector, paste the URL below and log in with your fal account when Claude asks. You're done when a new chat shows the fal tools.

Connector URL · fal MCPClick to copyhttps://mcp.fal.ai/mcp-relay

If your organisation restricts network access, allow fal.media in Settings → Capabilities.

03Give Claude access to a project folder

Claude Desktop

Open the Claude desktop app, link the chat to your computer and add the folder you want to work in. Claude can now read files in it and save results there. Claude only sees the folders you add.

04Describe your pipeline

Prompt

Start a new chat with the fal connector on and your folder linked. Tell Claude what you want to generate, in which order, and where to save it. The prompt below is a starting point, change the steps to your own project.

Example Prompt · Claude × falClick to copy

Use the fal connector for all generations and save everything into my linked project folder with clear file names. Before you start, tell me which model you'll use for each step and the estimated cost.

Pipeline:
1. Images: generate the stills I describe with an image model of your choice on fal. Save them to 02_IMAGES/ as NN_<name>.png and show them to me before moving on.
2. Video: animate the approved stills with FLUX 3 Max (image to video, endpoint blackforestlabs/flux-3-max/image-to-video), 5 seconds, 9:16, drafts at 720p. Save to 03_VIDEO/drafts/. Rerender at 1080p only when I approve a shot.
3. Cut: assemble the approved clips in order into one preview video with ffmpeg and save it to 04_EDIT/preview_v1.mp4.

Rules: log every prompt, seed and fal request ID in _prompts.md next to the files. Never regenerate everything, only what I ask for.

Note: Approve each stage before the next one runs. Later steps inherit earlier results.

05Refine in plain words

Iterate

Open the results and tell Claude what to change, like you'd brief an editor:

"Give me three versions of shot 04 with different seeds."
"Keep the drafts at 720p, we're still in the concept phase."
"Cut shot 02 shorter and add a slow push in."
"Switch the image model, this one struggles with hands."

Pro TipAsk Claude to keep a _prompts.md next to your files with every prompt, seed and fal request ID. Weeks later you can rerun any generation with the same settings.

Free Newsletter · 1,000+ Creators

Join the
Newsletter.

Get new AI tutorials, workflows and tool tips first, plus the free Image to Prompt Analyzer and the Top Creative AI Tools list.

Run AI image and video modelsfrom one chat.