Mage-Flow
Mage-Flow is Microsoft Research's MIT-licensed 4B image generation and editing family. The official model card provides base, reinforcement-learning-aligned, turbo, and editing variants for teams evaluating downloadable image models and custom serving.
Provider
Microsoft Research
Model family
Microsoft Mage
Open-weight image model
Cost tier
Flow
Status
Current
Release Jul 21, 2026
Why teams choose it
Use Mage-Flow when downloadable weights and variant control matter more than a managed
Microsoft image endpoint.
Benchmark the exact base, aligned, turbo, or editing checkpoint because speed and quality trade-offs differ.
Tradeoffs to know
- Self-hosting shifts safety, monitoring, capacity, and update responsibility to the deploying team.
- Review each checkpoint card and downstream dependencies before commercial deployment.
Benchmarks
No independent or vendor benchmark scores in this catalog entry.
Technical specs
- Inputs
- text, image
- Outputs
- image
- License
- MIT
- Parameter Count
- 4B family (official model card)
- Capabilities
- text-to-image
- image editing
- open-weight deployment
- turbo inference
- research fine-tuning
Related models
Mage-Flow FAQ
What is Mage-Flow?
Mage-Flow is Microsoft Research's MIT-licensed 4B image generation and editing family. The official model card provides base, reinforcement-learning-aligned, turbo, and editing variants for teams evaluating downloadable image models and custom serving.
When does Mage-Flow fit best?
Self-hosted image generation
What should teams watch out for with Mage-Flow?
Self-hosting shifts safety, monitoring, capacity, and update responsibility to the deploying team.
Explore next
Models, tools, and comparisons that connect to this reference.