Ollama FLUX.2 Klein Image Tool for Dify

A local Dify tool plugin that calls Ollama's OpenAI-compatible image generation endpoint and returns generated PNG files back to Dify.

What it does

• Calls POST /v1/images/generations on your Ollama server
• Uses x/flux2-klein by default
• Returns image files as Dify tool outputs
• Exposes workflow-friendly variables:
  • model
  • image_count
  • prompt_used

Important note

At the time this starter was prepared, Ollama's image-generation support for FLUX.2 Klein was documented as working on macOS, with Windows and Linux support still rolling out. If generation fails even though the plugin is wired correctly, verify your Ollama/image-model support first.

Python compatibility

Use Python 3.12 for local development and remote debug.

This plugin depends on dify_plugin, which currently requires tiktoken~=0.8.0. That tiktoken release provides wheels for Python 3.12 and 3.13, but not Python 3.14. On Python 3.14, pip falls back to building tiktoken from source, which requires a Rust toolchain.

If you see an error such as Building wheel for tiktoken ... error or can't find Rust compiler, switch to Python 3.12 instead of trying to install with your system python3.

Provider configuration

Recommended values:

• base_url
  • Remote debug from your local machine: http://localhost:11434
  • Plugin daemon in Docker talking to Ollama on the host: http://host.docker.internal:11434
• api_key
  • ollama
• default_model
  • x/flux2-klein
• timeout
  • 180

Tool parameters

• prompt - required
• size - optional, default 1024x1024
• n - optional, default 1
• model - optional override

Run in remote debug mode

1. Install the Dify CLI and make sure Python 3.12 is available.
2. Copy .env.example to .env.
3. Fill in your debug host, port, and key from Dify. REMOTE_INSTALL_KEY must be the real debug key from Dify. If you leave a placeholder such as your-debug-key, startup will fail with handshake failed, invalid key.
4. Create and activate a virtual environment with Python 3.12:

python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip

5. Install dependencies:

pip install -r requirements.txt

6. Start the plugin:

python -m main

If python3.12 is not installed on macOS, install it first:

brew install python@3.12

Package for local installation

From the parent directory:

dify plugin package ./ollama-flux2-klein-image-tool

This should create a .difypkg file that you can install from Dify using local file installation.

Example prompt

Create a premium Instagram launch visual for SKRILLEX with a dark luxury aesthetic, black and charcoal base, acid green accents, sharp editorial typography, cinematic lighting, centered focal point, strong negative space, high-end branding, include exact text: "SKRILLEX IS LIVE", "The next creative system starts now", "Discover More".