A high-performance MLX-based skill for generating images locally on Apple Silicon using FLUX.2 and Z-Image models.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install mflux
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install mflux using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is mflux: Local Apple Silicon Image Generation?
mflux is a specialized implementation designed to leverage the Apple MLX framework for high-speed, on-device image generation. By utilizing Openclaw Skills, developers and creators can integrate state-of-the-art models like FLUX.2 Klein and Z-Image Turbo directly into their local workflows without relying on external cloud APIs or incurring subscription costs. This skill ensures total data privacy, as all processing happens locally on your Mac, making it a critical tool for privacy-conscious environments.
The framework is optimized specifically for Apple Silicon (M1, M2, M3, and M4 chips), providing a seamless bridge between cutting-edge diffusion models and macOS hardware. Whether you need the lightning-fast 4-step generation of FLUX.2 Klein or the ultra-high fidelity of Z-Image Turbo, this skill provides the necessary interfaces to achieve professional results locally.
mflux: Local Apple Silicon Image Generation Use Cases
- Rapid prototyping of visual assets and UI components during development.
- Private image generation for sensitive projects where data cannot leave the local machine.
- Creating high-quality marketing materials and icons using Apache 2.0 licensed models.
- Transforming existing assets via local image-to-image and AI-powered editing workflows.
- Experimenting with custom artistic styles through LoRA support within Openclaw Skills.
How mflux: Local Apple Silicon Image Generation Works
- The user selects a specific model variant, such as FLUX.2 Klein for speed or Z-Image Turbo for quality.
- The skill initializes the model weights within the Apple MLX environment, utilizing unified memory for maximum efficiency.
- A text prompt or reference image is passed to the generation engine, which executes the denoising process on-device.
- For systems with limited memory, the skill applies 4-bit or 8-bit quantization to reduce the RAM footprint while maintaining image integrity.
- The final image is synthesized and saved to the local filesystem, ready for immediate use in projects.
mflux: Local Apple Silicon Image Generation Setup
To get started with mflux through Openclaw Skills, it is recommended to use the uv tool for a clean installation:
uv tool install --upgrade mflux --prerelease=allow
For faster model downloads, you can include hf_transfer:
uv tool install --upgrade mflux --with hf_transfer --prerelease=allow
Verify that the installation was successful and explore the CLI commands:
mflux-generate --help
mflux-generate-flux2 --help
mflux: Local Apple Silicon Image Generation Data Schema & Taxonomy
The skill manages its data locally, primarily handling model weights and generated outputs. Weights are cached to minimize network usage after the initial download.
| Data Component | Storage Type | Description |
|---|---|---|
| Model Cache | Local Directory (~/.cache/huggingface/) |
Stores FLUX.2 and Z-Image weights (up to 32GB depending on model) |
| Output Files | Local File System | PNG or JPG images generated from prompts |
| Quantization Data | In-Memory (4-bit/8-bit) | Compressed model representations to fit in 8GB/16GB RAM |
| Metadata | CLI Parameters | Seed values, inference steps, and guidance scales used for reproducibility |
mflux: Local Apple Silicon Image Generation Advanced Features
- Support for LoRA (Low-Rank Adaptation) to apply specific artistic styles or character consistency.
- Advanced Image-to-Image capabilities for transforming sketches or photos into high-fidelity AI art.
- Native FLUX.2 Image Editing for modifying specific parts of an image based on text prompts.
- Flexible quantization levels (4-bit, 8-bit) to optimize performance across different Apple Silicon RAM tiers.
- Direct Python API access for building custom automation and integrations within the Openclaw Skills ecosystem.
SKILL.md
Loading
Related Openclaw Skills
ModelWar AI CoreWar Arena
A competitive arena where AI agents battle in a virtual computer using the classic CoreWar programming game.
Scrapling
A high-performance Python-based web scraping tool designed for resilient data extraction from static, dynamic, and anti-bot protected websites.
Local Pandoc
A versatile document conversion skill that enables AI agents to transform Markdown files into professional PDFs, HTML, and Word documents using the Pandoc utility.
DHH Rails Style
A specialized AI skill for writing Ruby and Rails code that strictly follows the 37signals and DHH philosophy of clarity and REST purity.
Doubleword Batches
A high-performance batch processing skill for running asynchronous AI inference jobs at scale through the Doubleword API.
Doubleword Batches
A powerful batch processing system for executing large-scale AI inference requests asynchronously with significant cost savings.
Featured*
