4B parameters
The official release positions Mage-Flow as a compact 4B family, meaning it targets strong generation and editing quality without using tens of billions of parameters.
Prompt Gallery
Mage Flow AI prompt examples
Start with reusable Mage Flow AI prompt patterns for product shots, portraits, posters, bilingual text rendering, native wide crops, and precise image edits.
What is Mage-Flow
What is Mage-Flow?
Mage-Flow is Microsoft’s compact 4B image model family for text-to-image generation and instruction-based editing. In plain terms, it is built to make high-resolution image creation faster and lighter: 4B parameters, native 512-2048 output sizes, 4-step Turbo variants, and reported 1024x1024 latency of 0.59s for generation and 1.02s for editing on a single NVIDIA A100.
Interactive Turbo speed
The 4-step Turbo models are built for low-latency work. Microsoft reports 0.59s per 1024x1024 generation and 1.02s per edit on a single A100, with about 18-20 GB peak inference memory.
Native 512-2048 output
One checkpoint can generate flexible canvas sizes from 512 to 2048 pixels across many aspect ratios, including extreme 4:1 layouts such as 512x2048 or 2048x512.
Generation plus editing
Mage-Flow handles text-to-image prompts, while Mage-Flow-Edit supports semantic edits, appearance changes, restoration, object/content edits, and structure-aware outputs from a source image.
Mage Flow features
Mage-Flow features
The official Mage-Flow stack combines Mage-VAE, native-resolution packing, fused CUDA kernels, RL alignment, and few-step distillation. For prompt planning, that means output size, text rendering, style, and edit boundaries should be explicit before generation.
Model capability
Native-resolution image generation
Use detailed prompts for subject, scene, camera, style, layout, and constraints, then specify the output ratio so Mage Flow AI can target square, vertical, wide, or extreme native-resolution images.
Try it nowModel capability
Instruction-based image editing
Describe the exact region, object, background, or style that should change, and name what must stay untouched so Mage-Flow-Edit can preserve identity, composition, and the parts that already work.
Try it now
How it works
How to use a Mage Flow AI image generator workflow
Keep the Mage Flow AI loop simple: understand the model capability, write a concrete prompt, test a first result, then refine the prompt or edit instruction.
01
Choose generation or edit
Use text-to-image when you need a new visual direction. Use image edit when you already have a photo, product, portrait, room, or generated image to refine.
02
Set output size and aspect ratio
Mage-Flow is built for native-resolution output, so define whether the target is square, portrait, landscape, banner, or an extreme crop before writing style details.
03
Review the first output
Check whether the result followed the subject, style, text rendering, edit boundaries, and output format before changing the next prompt.
04
Refine with one clear change
Improve one thing at a time: lighting, background, camera, expression, product clarity, text legibility, crop, or realism. This keeps the Mage Flow AI workflow controllable.
Pricing
Mage Flow AI pricing
Start free, then use credits or plans for Mage Flow AI image generation, prompt testing, image editing, and commercial visual workflows.
50% OFF
Starter
$9.90/mo
$19.9
$118.80/YearSave 50%
100 credits ≈ $3.96
Perfect for trying out AI image generation and small projects.
- 250 credits per month (3000 credits/year)
- Yearly subscription credits are valid until your annual term ends
- Mage Flow AI workspace
- Mage Flow AI image generation
- Text-to-image workflows
- Reference image edits
- Private generations
- Commercial use
- No priority queue
- No team collaboration
Most Popular
50% OFF
Pro
$19.90/mo
$39.9
$238.80/YearSave 50%
100 credits ≈ $3.32
Ideal for creators, designers, and content professionals.
- 600 credits per month (7200 credits/year)
- Yearly subscription credits are valid until your annual term ends
- Mage Flow AI workspace
- Everything in Starter
- Higher generation throughput
- Priority queue
- Commercial use
- History saved
- Faster generation speed
- Best value per credit
40% OFF
Premium
$59.90/mo
$99.9
$718.80/YearSave 40%
100 credits ≈ $3.15
For teams, agencies, and high-volume commercial use.
- 1900 credits per month (22800 credits/year)
- Yearly subscription credits are valid until your annual term ends
- Mage Flow AI workspace
- Everything in Pro
- High-volume image production
- Fastest queue
- Commercial use
- Priority support
- Extended history
- Best for teams and agencies
FAQ
Mage Flow AI image generator questions
What is Mage-Flow?
Mage-Flow, also searched as mage-flow, is a Microsoft 4B-scale image generation and editing model family. Official materials describe it as a co-designed stack with Mage-VAE, a Native-Resolution Multimodal Diffusion Transformer, Base and RL-aligned checkpoints, plus 4-step Turbo variants.
Is Mage Flow an AI image generator?
Yes. The Mage Flow AI image generator is designed for text-to-image output, while Mage-Flow-Edit handles image editing from a source image plus an instruction.
What makes Mage-Flow different from larger image models?
The official paper emphasizes tokenizer, backbone, and system co-design. Mage-VAE reduces tokenization cost, and native-resolution packing helps one checkpoint work across many sizes and aspect ratios.
How fast is Mage-Flow-Turbo?
The paper reports that Mage-Flow-Turbo generates a 1024² image in 0.59 seconds on a single NVIDIA A100 GPU, while Mage-Flow-Edit-Turbo edits in 1.02 seconds.