Breaking: Docker Model Runner Now Powers Local AI Image Generation
In a major shift toward privacy-first AI, Docker today announced that its Model Runner can now generate images entirely on a user's local machine—no cloud subscriptions, no data leaks, and no content filters.
The new capability pairs Docker Model Runner with Open WebUI, a popular open-source chat interface, to deliver a fully private, on-premises alternative to services like DALL-E and Midjourney.
Users can pull a model, launch a web UI, and start creating images—all from a few terminal commands.
Key Features at a Glance
- Complete privacy: All prompts and generated images stay on your hardware.
- No recurring costs: No credit-based billing or subscription fees.
- OpenAI-compatible API: Works with any tool that supports
/v1/images/generations. - Minimal hardware requirements: 8 GB of RAM and optional GPU acceleration (NVIDIA CUDA, Apple Silicon MPS, or CPU fallback).
How It Works
Docker Model Runner acts as the control plane. It downloads the model using a new packaging format called DDUF (Diffusers Unified Format), manages the inference backend, and exposes a fully OpenAI-compatible API endpoint.
Open WebUI connects to that endpoint automatically, providing a chat-based interface for generating images.
"This is a game-changer for developers and designers who need to iterate on visual content without worrying about privacy or costs," said Clara Williams, Docker's Director of Product Management. "You essentially get your own private DALL-E, running on your laptop."
Getting Started in Two Commands
To pull an image generation model, users run:
docker model pull stable-diffusionThen launch the web UI with:
docker model launch openwebuiThat's it. The model is stored locally as a DDUF file—a single artifact bundling all diffusion components (text encoder, VAE, UNet, scheduler config). Docker Model Runner unpacks it at runtime.
Background: The Problem with Cloud-Based Image Generation
Until now, most users relied on cloud services to generate AI images. This meant sending prompts to remote servers, paying per generation, and accepting arbitrary content filters that often blocked legitimate requests—such as "a dragon wearing a business suit."
Privacy concerns also loomed large: prompts and generated images could be stored, analyzed, or used for training. For companies handling sensitive data, this was unacceptable.
Docker's solution eliminates these trade-offs by running everything locally. The open-source Open WebUI provides a polished interface, while Docker Model Runner handles the heavy lifting of model distribution and execution.
What This Means for Users
Professionals in design, marketing, and software development can now generate unlimited images without budget constraints or privacy risks. Small teams and independent creators gain access to state-of-the-art AI without vendor lock-in.
The DDUF format also simplifies model distribution—no more complex setups or missing dependencies. As more models adopt this format, users will have a growing library of locally runnable AI tools.
"We see this as the first step toward a fully offline AI ecosystem," added Williams. "Image generation is just the beginning."
Requirements and Next Steps
Users need Docker Desktop (macOS) or Docker Engine (Linux), about 8 GB of RAM, and optionally a GPU. The initial model, Stable Diffusion XL, is 6.94 GB and available via docker model pull stable-diffusion.
Docker plans to support more models and features, including advanced fine-tuning and control over generation parameters in future releases.