Open-source 4B image generation and editing model with built-in ControlNet-style conditioning and a 4-step turbo variant. Self-reported benchmarks are close to Flux Kontext and Qwen Image; main model is 17.5 GB.
Mage Flow
by Microsoft
Microsoft's open-weights 4B model for image generation and editing, with a 4-step turbo variant.
Updated EU status checked Last shown
Mage Flow at a glance
- What is Mage Flow?
- Mage Flow is a model from Microsoft. Microsoft's open-weights 4B model for image generation and editing, with a 4-step turbo variant.
- Who makes Mage Flow?
- Mage Flow is made by Microsoft.
- Does Mage Flow work in the EU?
- Available in the EU. Open weights are downloadable from Hugging Face without regional gating, so they can be run in the EU. No license text was fetched, so no EU exclusion was confirmed or ruled out. Last verified 14 Nov 2026.
- How much does Mage Flow cost?
- Open source. Weights are free to download from Hugging Face; local compute costs are the user's own.
- Where was Mage Flow shown?
- In 1 creator video by AI Search. Most recently by AI Search on 26 Jul 2026, at 1:05 in “Claude Opus 5, GPT 6 hack, Flux 3, new Gemini, quantum breakthrough, new Qwen: AI NEWS”.
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Mage Flow is an image generation and editing model published by Microsoft on Hugging Face. According to a YouTube creator review, it has about 4B parameters, with a main checkpoint of roughly 17.5 GB, and includes built-in ControlNet-style conditioning, so structural guidance does not need a separate add-on network.
It supports text-to-image generation and instruction-based image editing. A 4-step turbo variant is available for faster inference. Self-reported benchmarks are described as close to Flux Kontext and Qwen Image; these figures have not been independently verified here.
The weights are hosted at huggingface.co/microsoft/Mage-Flow for local use. The fetched pages did not give license terms, a paper, or a release date, so those details should be checked on the model card.
- About 4B parameters
- Image generation and image editing in one model
- Built-in ControlNet-style conditioning
- 4-step turbo variant for faster inference
- Weights downloadable from Hugging Face
As seen on YouTube
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Similar AI tools
Closest neighbours by meaning, not just by category.
Qwen Image 2.1 Turbo
Model · Alibaba
Open-weight accelerated variant of Alibaba's Qwen Image 2.1 for faster text-to-image generation and image editing.
MOSS-Image
Model · OpenMOSS
Open-weights 5B text-to-image model family from OpenMOSS with native JSON bounding-box layout control and text rendering.
SenseNova U1.5 8B
Model · SenseTime
Open-weights 8B image generation and editing model from SenseTime that works in pixel space without a VAE.
GPT Image 2.6
Model · OpenAI
OpenAI image generation and editing model with GIF output, SVG export and image markup, in Flare and Sunburst variants.
Krea 2 Turbo
Model · Krea
Faster variant of Krea's Krea 2 image model, reported to generate images in about 2 seconds.
LLaDA-Image
Model · inclusionAI
Open-weights 6B unified text-to-image and instruction-based image editing model family from inclusionAI.